xref: /linux/drivers/net/ethernet/hisilicon/hns3/hns3_enet.c (revision 9c39c6ffe0c2945c7cf814814c096bc23b63f53d)
1 // SPDX-License-Identifier: GPL-2.0+
2 // Copyright (c) 2016-2017 Hisilicon Limited.
3 
4 #include <linux/dma-mapping.h>
5 #include <linux/etherdevice.h>
6 #include <linux/interrupt.h>
7 #ifdef CONFIG_RFS_ACCEL
8 #include <linux/cpu_rmap.h>
9 #endif
10 #include <linux/if_vlan.h>
11 #include <linux/irq.h>
12 #include <linux/ip.h>
13 #include <linux/ipv6.h>
14 #include <linux/module.h>
15 #include <linux/pci.h>
16 #include <linux/aer.h>
17 #include <linux/skbuff.h>
18 #include <linux/sctp.h>
19 #include <net/gre.h>
20 #include <net/ip6_checksum.h>
21 #include <net/pkt_cls.h>
22 #include <net/tcp.h>
23 #include <net/vxlan.h>
24 #include <net/geneve.h>
25 
26 #include "hnae3.h"
27 #include "hns3_enet.h"
28 /* All hns3 tracepoints are defined by the include below, which
29  * must be included exactly once across the whole kernel with
30  * CREATE_TRACE_POINTS defined
31  */
32 #define CREATE_TRACE_POINTS
33 #include "hns3_trace.h"
34 
35 #define hns3_set_field(origin, shift, val)	((origin) |= (val) << (shift))
36 #define hns3_tx_bd_count(S)	DIV_ROUND_UP(S, HNS3_MAX_BD_SIZE)
37 
38 #define hns3_rl_err(fmt, ...)						\
39 	do {								\
40 		if (net_ratelimit())					\
41 			netdev_err(fmt, ##__VA_ARGS__);			\
42 	} while (0)
43 
44 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force);
45 
46 static const char hns3_driver_name[] = "hns3";
47 static const char hns3_driver_string[] =
48 			"Hisilicon Ethernet Network Driver for Hip08 Family";
49 static const char hns3_copyright[] = "Copyright (c) 2017 Huawei Corporation.";
50 static struct hnae3_client client;
51 
52 static int debug = -1;
53 module_param(debug, int, 0);
54 MODULE_PARM_DESC(debug, " Network interface message level setting");
55 
56 #define DEFAULT_MSG_LEVEL (NETIF_MSG_PROBE | NETIF_MSG_LINK | \
57 			   NETIF_MSG_IFDOWN | NETIF_MSG_IFUP)
58 
59 #define HNS3_INNER_VLAN_TAG	1
60 #define HNS3_OUTER_VLAN_TAG	2
61 
62 #define HNS3_MIN_TX_LEN		33U
63 
64 /* hns3_pci_tbl - PCI Device ID Table
65  *
66  * Last entry must be all 0s
67  *
68  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
69  *   Class, Class Mask, private data (not used) }
70  */
71 static const struct pci_device_id hns3_pci_tbl[] = {
72 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
73 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
74 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA),
75 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
76 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC),
77 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
78 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA),
79 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
80 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC),
81 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
82 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC),
83 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
84 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA),
85 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
86 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_VF), 0},
87 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
88 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
89 	/* required last entry */
90 	{0, }
91 };
92 MODULE_DEVICE_TABLE(pci, hns3_pci_tbl);
93 
94 static irqreturn_t hns3_irq_handle(int irq, void *vector)
95 {
96 	struct hns3_enet_tqp_vector *tqp_vector = vector;
97 
98 	napi_schedule_irqoff(&tqp_vector->napi);
99 
100 	return IRQ_HANDLED;
101 }
102 
103 static void hns3_nic_uninit_irq(struct hns3_nic_priv *priv)
104 {
105 	struct hns3_enet_tqp_vector *tqp_vectors;
106 	unsigned int i;
107 
108 	for (i = 0; i < priv->vector_num; i++) {
109 		tqp_vectors = &priv->tqp_vector[i];
110 
111 		if (tqp_vectors->irq_init_flag != HNS3_VECTOR_INITED)
112 			continue;
113 
114 		/* clear the affinity mask */
115 		irq_set_affinity_hint(tqp_vectors->vector_irq, NULL);
116 
117 		/* release the irq resource */
118 		free_irq(tqp_vectors->vector_irq, tqp_vectors);
119 		tqp_vectors->irq_init_flag = HNS3_VECTOR_NOT_INITED;
120 	}
121 }
122 
123 static int hns3_nic_init_irq(struct hns3_nic_priv *priv)
124 {
125 	struct hns3_enet_tqp_vector *tqp_vectors;
126 	int txrx_int_idx = 0;
127 	int rx_int_idx = 0;
128 	int tx_int_idx = 0;
129 	unsigned int i;
130 	int ret;
131 
132 	for (i = 0; i < priv->vector_num; i++) {
133 		tqp_vectors = &priv->tqp_vector[i];
134 
135 		if (tqp_vectors->irq_init_flag == HNS3_VECTOR_INITED)
136 			continue;
137 
138 		if (tqp_vectors->tx_group.ring && tqp_vectors->rx_group.ring) {
139 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
140 				 "%s-%s-%s-%d", hns3_driver_name,
141 				 pci_name(priv->ae_handle->pdev),
142 				 "TxRx", txrx_int_idx++);
143 			txrx_int_idx++;
144 		} else if (tqp_vectors->rx_group.ring) {
145 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
146 				 "%s-%s-%s-%d", hns3_driver_name,
147 				 pci_name(priv->ae_handle->pdev),
148 				 "Rx", rx_int_idx++);
149 		} else if (tqp_vectors->tx_group.ring) {
150 			snprintf(tqp_vectors->name, HNAE3_INT_NAME_LEN,
151 				 "%s-%s-%s-%d", hns3_driver_name,
152 				 pci_name(priv->ae_handle->pdev),
153 				 "Tx", tx_int_idx++);
154 		} else {
155 			/* Skip this unused q_vector */
156 			continue;
157 		}
158 
159 		tqp_vectors->name[HNAE3_INT_NAME_LEN - 1] = '\0';
160 
161 		irq_set_status_flags(tqp_vectors->vector_irq, IRQ_NOAUTOEN);
162 		ret = request_irq(tqp_vectors->vector_irq, hns3_irq_handle, 0,
163 				  tqp_vectors->name, tqp_vectors);
164 		if (ret) {
165 			netdev_err(priv->netdev, "request irq(%d) fail\n",
166 				   tqp_vectors->vector_irq);
167 			hns3_nic_uninit_irq(priv);
168 			return ret;
169 		}
170 
171 		irq_set_affinity_hint(tqp_vectors->vector_irq,
172 				      &tqp_vectors->affinity_mask);
173 
174 		tqp_vectors->irq_init_flag = HNS3_VECTOR_INITED;
175 	}
176 
177 	return 0;
178 }
179 
180 static void hns3_mask_vector_irq(struct hns3_enet_tqp_vector *tqp_vector,
181 				 u32 mask_en)
182 {
183 	writel(mask_en, tqp_vector->mask_addr);
184 }
185 
186 static void hns3_vector_enable(struct hns3_enet_tqp_vector *tqp_vector)
187 {
188 	napi_enable(&tqp_vector->napi);
189 	enable_irq(tqp_vector->vector_irq);
190 
191 	/* enable vector */
192 	hns3_mask_vector_irq(tqp_vector, 1);
193 }
194 
195 static void hns3_vector_disable(struct hns3_enet_tqp_vector *tqp_vector)
196 {
197 	/* disable vector */
198 	hns3_mask_vector_irq(tqp_vector, 0);
199 
200 	disable_irq(tqp_vector->vector_irq);
201 	napi_disable(&tqp_vector->napi);
202 }
203 
204 void hns3_set_vector_coalesce_rl(struct hns3_enet_tqp_vector *tqp_vector,
205 				 u32 rl_value)
206 {
207 	u32 rl_reg = hns3_rl_usec_to_reg(rl_value);
208 
209 	/* this defines the configuration for RL (Interrupt Rate Limiter).
210 	 * Rl defines rate of interrupts i.e. number of interrupts-per-second
211 	 * GL and RL(Rate Limiter) are 2 ways to acheive interrupt coalescing
212 	 */
213 	if (rl_reg > 0 && !tqp_vector->tx_group.coal.adapt_enable &&
214 	    !tqp_vector->rx_group.coal.adapt_enable)
215 		/* According to the hardware, the range of rl_reg is
216 		 * 0-59 and the unit is 4.
217 		 */
218 		rl_reg |=  HNS3_INT_RL_ENABLE_MASK;
219 
220 	writel(rl_reg, tqp_vector->mask_addr + HNS3_VECTOR_RL_OFFSET);
221 }
222 
223 void hns3_set_vector_coalesce_rx_gl(struct hns3_enet_tqp_vector *tqp_vector,
224 				    u32 gl_value)
225 {
226 	u32 new_val;
227 
228 	if (tqp_vector->rx_group.coal.unit_1us)
229 		new_val = gl_value | HNS3_INT_GL_1US;
230 	else
231 		new_val = hns3_gl_usec_to_reg(gl_value);
232 
233 	writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL0_OFFSET);
234 }
235 
236 void hns3_set_vector_coalesce_tx_gl(struct hns3_enet_tqp_vector *tqp_vector,
237 				    u32 gl_value)
238 {
239 	u32 new_val;
240 
241 	if (tqp_vector->tx_group.coal.unit_1us)
242 		new_val = gl_value | HNS3_INT_GL_1US;
243 	else
244 		new_val = hns3_gl_usec_to_reg(gl_value);
245 
246 	writel(new_val, tqp_vector->mask_addr + HNS3_VECTOR_GL1_OFFSET);
247 }
248 
249 void hns3_set_vector_coalesce_tx_ql(struct hns3_enet_tqp_vector *tqp_vector,
250 				    u32 ql_value)
251 {
252 	writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_TX_QL_OFFSET);
253 }
254 
255 void hns3_set_vector_coalesce_rx_ql(struct hns3_enet_tqp_vector *tqp_vector,
256 				    u32 ql_value)
257 {
258 	writel(ql_value, tqp_vector->mask_addr + HNS3_VECTOR_RX_QL_OFFSET);
259 }
260 
261 static void hns3_vector_coalesce_init(struct hns3_enet_tqp_vector *tqp_vector,
262 				      struct hns3_nic_priv *priv)
263 {
264 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
265 	struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
266 	struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
267 
268 	/* initialize the configuration for interrupt coalescing.
269 	 * 1. GL (Interrupt Gap Limiter)
270 	 * 2. RL (Interrupt Rate Limiter)
271 	 * 3. QL (Interrupt Quantity Limiter)
272 	 *
273 	 * Default: enable interrupt coalescing self-adaptive and GL
274 	 */
275 	tx_coal->adapt_enable = 1;
276 	rx_coal->adapt_enable = 1;
277 
278 	tx_coal->int_gl = HNS3_INT_GL_50K;
279 	rx_coal->int_gl = HNS3_INT_GL_50K;
280 
281 	rx_coal->flow_level = HNS3_FLOW_LOW;
282 	tx_coal->flow_level = HNS3_FLOW_LOW;
283 
284 	/* device version above V3(include V3), GL can configure 1us
285 	 * unit, so uses 1us unit.
286 	 */
287 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3) {
288 		tx_coal->unit_1us = 1;
289 		rx_coal->unit_1us = 1;
290 	}
291 
292 	if (ae_dev->dev_specs.int_ql_max) {
293 		tx_coal->ql_enable = 1;
294 		rx_coal->ql_enable = 1;
295 		tx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
296 		rx_coal->int_ql_max = ae_dev->dev_specs.int_ql_max;
297 		tx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
298 		rx_coal->int_ql = HNS3_INT_QL_DEFAULT_CFG;
299 	}
300 }
301 
302 static void
303 hns3_vector_coalesce_init_hw(struct hns3_enet_tqp_vector *tqp_vector,
304 			     struct hns3_nic_priv *priv)
305 {
306 	struct hns3_enet_coalesce *tx_coal = &tqp_vector->tx_group.coal;
307 	struct hns3_enet_coalesce *rx_coal = &tqp_vector->rx_group.coal;
308 	struct hnae3_handle *h = priv->ae_handle;
309 
310 	hns3_set_vector_coalesce_tx_gl(tqp_vector, tx_coal->int_gl);
311 	hns3_set_vector_coalesce_rx_gl(tqp_vector, rx_coal->int_gl);
312 	hns3_set_vector_coalesce_rl(tqp_vector, h->kinfo.int_rl_setting);
313 
314 	if (tx_coal->ql_enable)
315 		hns3_set_vector_coalesce_tx_ql(tqp_vector, tx_coal->int_ql);
316 
317 	if (rx_coal->ql_enable)
318 		hns3_set_vector_coalesce_rx_ql(tqp_vector, rx_coal->int_ql);
319 }
320 
321 static int hns3_nic_set_real_num_queue(struct net_device *netdev)
322 {
323 	struct hnae3_handle *h = hns3_get_handle(netdev);
324 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
325 	struct hnae3_tc_info *tc_info = &kinfo->tc_info;
326 	unsigned int queue_size = kinfo->num_tqps;
327 	int i, ret;
328 
329 	if (tc_info->num_tc <= 1 && !tc_info->mqprio_active) {
330 		netdev_reset_tc(netdev);
331 	} else {
332 		ret = netdev_set_num_tc(netdev, tc_info->num_tc);
333 		if (ret) {
334 			netdev_err(netdev,
335 				   "netdev_set_num_tc fail, ret=%d!\n", ret);
336 			return ret;
337 		}
338 
339 		for (i = 0; i < HNAE3_MAX_TC; i++) {
340 			if (!test_bit(i, &tc_info->tc_en))
341 				continue;
342 
343 			netdev_set_tc_queue(netdev, i, tc_info->tqp_count[i],
344 					    tc_info->tqp_offset[i]);
345 		}
346 	}
347 
348 	ret = netif_set_real_num_tx_queues(netdev, queue_size);
349 	if (ret) {
350 		netdev_err(netdev,
351 			   "netif_set_real_num_tx_queues fail, ret=%d!\n", ret);
352 		return ret;
353 	}
354 
355 	ret = netif_set_real_num_rx_queues(netdev, queue_size);
356 	if (ret) {
357 		netdev_err(netdev,
358 			   "netif_set_real_num_rx_queues fail, ret=%d!\n", ret);
359 		return ret;
360 	}
361 
362 	return 0;
363 }
364 
365 static u16 hns3_get_max_available_channels(struct hnae3_handle *h)
366 {
367 	u16 alloc_tqps, max_rss_size, rss_size;
368 
369 	h->ae_algo->ops->get_tqps_and_rss_info(h, &alloc_tqps, &max_rss_size);
370 	rss_size = alloc_tqps / h->kinfo.tc_info.num_tc;
371 
372 	return min_t(u16, rss_size, max_rss_size);
373 }
374 
375 static void hns3_tqp_enable(struct hnae3_queue *tqp)
376 {
377 	u32 rcb_reg;
378 
379 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
380 	rcb_reg |= BIT(HNS3_RING_EN_B);
381 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
382 }
383 
384 static void hns3_tqp_disable(struct hnae3_queue *tqp)
385 {
386 	u32 rcb_reg;
387 
388 	rcb_reg = hns3_read_dev(tqp, HNS3_RING_EN_REG);
389 	rcb_reg &= ~BIT(HNS3_RING_EN_B);
390 	hns3_write_dev(tqp, HNS3_RING_EN_REG, rcb_reg);
391 }
392 
393 static void hns3_free_rx_cpu_rmap(struct net_device *netdev)
394 {
395 #ifdef CONFIG_RFS_ACCEL
396 	free_irq_cpu_rmap(netdev->rx_cpu_rmap);
397 	netdev->rx_cpu_rmap = NULL;
398 #endif
399 }
400 
401 static int hns3_set_rx_cpu_rmap(struct net_device *netdev)
402 {
403 #ifdef CONFIG_RFS_ACCEL
404 	struct hns3_nic_priv *priv = netdev_priv(netdev);
405 	struct hns3_enet_tqp_vector *tqp_vector;
406 	int i, ret;
407 
408 	if (!netdev->rx_cpu_rmap) {
409 		netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->vector_num);
410 		if (!netdev->rx_cpu_rmap)
411 			return -ENOMEM;
412 	}
413 
414 	for (i = 0; i < priv->vector_num; i++) {
415 		tqp_vector = &priv->tqp_vector[i];
416 		ret = irq_cpu_rmap_add(netdev->rx_cpu_rmap,
417 				       tqp_vector->vector_irq);
418 		if (ret) {
419 			hns3_free_rx_cpu_rmap(netdev);
420 			return ret;
421 		}
422 	}
423 #endif
424 	return 0;
425 }
426 
427 static int hns3_nic_net_up(struct net_device *netdev)
428 {
429 	struct hns3_nic_priv *priv = netdev_priv(netdev);
430 	struct hnae3_handle *h = priv->ae_handle;
431 	int i, j;
432 	int ret;
433 
434 	ret = hns3_nic_reset_all_ring(h);
435 	if (ret)
436 		return ret;
437 
438 	clear_bit(HNS3_NIC_STATE_DOWN, &priv->state);
439 
440 	/* enable the vectors */
441 	for (i = 0; i < priv->vector_num; i++)
442 		hns3_vector_enable(&priv->tqp_vector[i]);
443 
444 	/* enable rcb */
445 	for (j = 0; j < h->kinfo.num_tqps; j++)
446 		hns3_tqp_enable(h->kinfo.tqp[j]);
447 
448 	/* start the ae_dev */
449 	ret = h->ae_algo->ops->start ? h->ae_algo->ops->start(h) : 0;
450 	if (ret) {
451 		set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
452 		while (j--)
453 			hns3_tqp_disable(h->kinfo.tqp[j]);
454 
455 		for (j = i - 1; j >= 0; j--)
456 			hns3_vector_disable(&priv->tqp_vector[j]);
457 	}
458 
459 	return ret;
460 }
461 
462 static void hns3_config_xps(struct hns3_nic_priv *priv)
463 {
464 	int i;
465 
466 	for (i = 0; i < priv->vector_num; i++) {
467 		struct hns3_enet_tqp_vector *tqp_vector = &priv->tqp_vector[i];
468 		struct hns3_enet_ring *ring = tqp_vector->tx_group.ring;
469 
470 		while (ring) {
471 			int ret;
472 
473 			ret = netif_set_xps_queue(priv->netdev,
474 						  &tqp_vector->affinity_mask,
475 						  ring->tqp->tqp_index);
476 			if (ret)
477 				netdev_warn(priv->netdev,
478 					    "set xps queue failed: %d", ret);
479 
480 			ring = ring->next;
481 		}
482 	}
483 }
484 
485 static int hns3_nic_net_open(struct net_device *netdev)
486 {
487 	struct hns3_nic_priv *priv = netdev_priv(netdev);
488 	struct hnae3_handle *h = hns3_get_handle(netdev);
489 	struct hnae3_knic_private_info *kinfo;
490 	int i, ret;
491 
492 	if (hns3_nic_resetting(netdev))
493 		return -EBUSY;
494 
495 	netif_carrier_off(netdev);
496 
497 	ret = hns3_nic_set_real_num_queue(netdev);
498 	if (ret)
499 		return ret;
500 
501 	ret = hns3_nic_net_up(netdev);
502 	if (ret) {
503 		netdev_err(netdev, "net up fail, ret=%d!\n", ret);
504 		return ret;
505 	}
506 
507 	kinfo = &h->kinfo;
508 	for (i = 0; i < HNAE3_MAX_USER_PRIO; i++)
509 		netdev_set_prio_tc_map(netdev, i, kinfo->tc_info.prio_tc[i]);
510 
511 	if (h->ae_algo->ops->set_timer_task)
512 		h->ae_algo->ops->set_timer_task(priv->ae_handle, true);
513 
514 	hns3_config_xps(priv);
515 
516 	netif_dbg(h, drv, netdev, "net open\n");
517 
518 	return 0;
519 }
520 
521 static void hns3_reset_tx_queue(struct hnae3_handle *h)
522 {
523 	struct net_device *ndev = h->kinfo.netdev;
524 	struct hns3_nic_priv *priv = netdev_priv(ndev);
525 	struct netdev_queue *dev_queue;
526 	u32 i;
527 
528 	for (i = 0; i < h->kinfo.num_tqps; i++) {
529 		dev_queue = netdev_get_tx_queue(ndev,
530 						priv->ring[i].queue_index);
531 		netdev_tx_reset_queue(dev_queue);
532 	}
533 }
534 
535 static void hns3_nic_net_down(struct net_device *netdev)
536 {
537 	struct hns3_nic_priv *priv = netdev_priv(netdev);
538 	struct hnae3_handle *h = hns3_get_handle(netdev);
539 	const struct hnae3_ae_ops *ops;
540 	int i;
541 
542 	/* disable vectors */
543 	for (i = 0; i < priv->vector_num; i++)
544 		hns3_vector_disable(&priv->tqp_vector[i]);
545 
546 	/* disable rcb */
547 	for (i = 0; i < h->kinfo.num_tqps; i++)
548 		hns3_tqp_disable(h->kinfo.tqp[i]);
549 
550 	/* stop ae_dev */
551 	ops = priv->ae_handle->ae_algo->ops;
552 	if (ops->stop)
553 		ops->stop(priv->ae_handle);
554 
555 	/* delay ring buffer clearing to hns3_reset_notify_uninit_enet
556 	 * during reset process, because driver may not be able
557 	 * to disable the ring through firmware when downing the netdev.
558 	 */
559 	if (!hns3_nic_resetting(netdev))
560 		hns3_clear_all_ring(priv->ae_handle, false);
561 
562 	hns3_reset_tx_queue(priv->ae_handle);
563 }
564 
565 static int hns3_nic_net_stop(struct net_device *netdev)
566 {
567 	struct hns3_nic_priv *priv = netdev_priv(netdev);
568 	struct hnae3_handle *h = hns3_get_handle(netdev);
569 
570 	if (test_and_set_bit(HNS3_NIC_STATE_DOWN, &priv->state))
571 		return 0;
572 
573 	netif_dbg(h, drv, netdev, "net stop\n");
574 
575 	if (h->ae_algo->ops->set_timer_task)
576 		h->ae_algo->ops->set_timer_task(priv->ae_handle, false);
577 
578 	netif_tx_stop_all_queues(netdev);
579 	netif_carrier_off(netdev);
580 
581 	hns3_nic_net_down(netdev);
582 
583 	return 0;
584 }
585 
586 static int hns3_nic_uc_sync(struct net_device *netdev,
587 			    const unsigned char *addr)
588 {
589 	struct hnae3_handle *h = hns3_get_handle(netdev);
590 
591 	if (h->ae_algo->ops->add_uc_addr)
592 		return h->ae_algo->ops->add_uc_addr(h, addr);
593 
594 	return 0;
595 }
596 
597 static int hns3_nic_uc_unsync(struct net_device *netdev,
598 			      const unsigned char *addr)
599 {
600 	struct hnae3_handle *h = hns3_get_handle(netdev);
601 
602 	/* need ignore the request of removing device address, because
603 	 * we store the device address and other addresses of uc list
604 	 * in the function's mac filter list.
605 	 */
606 	if (ether_addr_equal(addr, netdev->dev_addr))
607 		return 0;
608 
609 	if (h->ae_algo->ops->rm_uc_addr)
610 		return h->ae_algo->ops->rm_uc_addr(h, addr);
611 
612 	return 0;
613 }
614 
615 static int hns3_nic_mc_sync(struct net_device *netdev,
616 			    const unsigned char *addr)
617 {
618 	struct hnae3_handle *h = hns3_get_handle(netdev);
619 
620 	if (h->ae_algo->ops->add_mc_addr)
621 		return h->ae_algo->ops->add_mc_addr(h, addr);
622 
623 	return 0;
624 }
625 
626 static int hns3_nic_mc_unsync(struct net_device *netdev,
627 			      const unsigned char *addr)
628 {
629 	struct hnae3_handle *h = hns3_get_handle(netdev);
630 
631 	if (h->ae_algo->ops->rm_mc_addr)
632 		return h->ae_algo->ops->rm_mc_addr(h, addr);
633 
634 	return 0;
635 }
636 
637 static u8 hns3_get_netdev_flags(struct net_device *netdev)
638 {
639 	u8 flags = 0;
640 
641 	if (netdev->flags & IFF_PROMISC) {
642 		flags = HNAE3_USER_UPE | HNAE3_USER_MPE | HNAE3_BPE;
643 	} else {
644 		flags |= HNAE3_VLAN_FLTR;
645 		if (netdev->flags & IFF_ALLMULTI)
646 			flags |= HNAE3_USER_MPE;
647 	}
648 
649 	return flags;
650 }
651 
652 static void hns3_nic_set_rx_mode(struct net_device *netdev)
653 {
654 	struct hnae3_handle *h = hns3_get_handle(netdev);
655 	u8 new_flags;
656 
657 	new_flags = hns3_get_netdev_flags(netdev);
658 
659 	__dev_uc_sync(netdev, hns3_nic_uc_sync, hns3_nic_uc_unsync);
660 	__dev_mc_sync(netdev, hns3_nic_mc_sync, hns3_nic_mc_unsync);
661 
662 	/* User mode Promisc mode enable and vlan filtering is disabled to
663 	 * let all packets in.
664 	 */
665 	h->netdev_flags = new_flags;
666 	hns3_request_update_promisc_mode(h);
667 }
668 
669 void hns3_request_update_promisc_mode(struct hnae3_handle *handle)
670 {
671 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
672 
673 	if (ops->request_update_promisc_mode)
674 		ops->request_update_promisc_mode(handle);
675 }
676 
677 void hns3_enable_vlan_filter(struct net_device *netdev, bool enable)
678 {
679 	struct hns3_nic_priv *priv = netdev_priv(netdev);
680 	struct hnae3_handle *h = priv->ae_handle;
681 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(h->pdev);
682 	bool last_state;
683 
684 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2 &&
685 	    h->ae_algo->ops->enable_vlan_filter) {
686 		last_state = h->netdev_flags & HNAE3_VLAN_FLTR ? true : false;
687 		if (enable != last_state) {
688 			netdev_info(netdev,
689 				    "%s vlan filter\n",
690 				    enable ? "enable" : "disable");
691 			h->ae_algo->ops->enable_vlan_filter(h, enable);
692 		}
693 	}
694 }
695 
696 static int hns3_set_tso(struct sk_buff *skb, u32 *paylen_fdop_ol4cs,
697 			u16 *mss, u32 *type_cs_vlan_tso, u32 *send_bytes)
698 {
699 	u32 l4_offset, hdr_len;
700 	union l3_hdr_info l3;
701 	union l4_hdr_info l4;
702 	u32 l4_paylen;
703 	int ret;
704 
705 	if (!skb_is_gso(skb))
706 		return 0;
707 
708 	ret = skb_cow_head(skb, 0);
709 	if (unlikely(ret < 0))
710 		return ret;
711 
712 	l3.hdr = skb_network_header(skb);
713 	l4.hdr = skb_transport_header(skb);
714 
715 	/* Software should clear the IPv4's checksum field when tso is
716 	 * needed.
717 	 */
718 	if (l3.v4->version == 4)
719 		l3.v4->check = 0;
720 
721 	/* tunnel packet */
722 	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
723 					 SKB_GSO_GRE_CSUM |
724 					 SKB_GSO_UDP_TUNNEL |
725 					 SKB_GSO_UDP_TUNNEL_CSUM)) {
726 		/* reset l3&l4 pointers from outer to inner headers */
727 		l3.hdr = skb_inner_network_header(skb);
728 		l4.hdr = skb_inner_transport_header(skb);
729 
730 		/* Software should clear the IPv4's checksum field when
731 		 * tso is needed.
732 		 */
733 		if (l3.v4->version == 4)
734 			l3.v4->check = 0;
735 	}
736 
737 	/* normal or tunnel packet */
738 	l4_offset = l4.hdr - skb->data;
739 
740 	/* remove payload length from inner pseudo checksum when tso */
741 	l4_paylen = skb->len - l4_offset;
742 
743 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
744 		hdr_len = sizeof(*l4.udp) + l4_offset;
745 		csum_replace_by_diff(&l4.udp->check,
746 				     (__force __wsum)htonl(l4_paylen));
747 	} else {
748 		hdr_len = (l4.tcp->doff << 2) + l4_offset;
749 		csum_replace_by_diff(&l4.tcp->check,
750 				     (__force __wsum)htonl(l4_paylen));
751 	}
752 
753 	*send_bytes = (skb_shinfo(skb)->gso_segs - 1) * hdr_len + skb->len;
754 
755 	/* find the txbd field values */
756 	*paylen_fdop_ol4cs = skb->len - hdr_len;
757 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_TSO_B, 1);
758 
759 	/* offload outer UDP header checksum */
760 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)
761 		hns3_set_field(*paylen_fdop_ol4cs, HNS3_TXD_OL4CS_B, 1);
762 
763 	/* get MSS for TSO */
764 	*mss = skb_shinfo(skb)->gso_size;
765 
766 	trace_hns3_tso(skb);
767 
768 	return 0;
769 }
770 
771 static int hns3_get_l4_protocol(struct sk_buff *skb, u8 *ol4_proto,
772 				u8 *il4_proto)
773 {
774 	union l3_hdr_info l3;
775 	unsigned char *l4_hdr;
776 	unsigned char *exthdr;
777 	u8 l4_proto_tmp;
778 	__be16 frag_off;
779 
780 	/* find outer header point */
781 	l3.hdr = skb_network_header(skb);
782 	l4_hdr = skb_transport_header(skb);
783 
784 	if (skb->protocol == htons(ETH_P_IPV6)) {
785 		exthdr = l3.hdr + sizeof(*l3.v6);
786 		l4_proto_tmp = l3.v6->nexthdr;
787 		if (l4_hdr != exthdr)
788 			ipv6_skip_exthdr(skb, exthdr - skb->data,
789 					 &l4_proto_tmp, &frag_off);
790 	} else if (skb->protocol == htons(ETH_P_IP)) {
791 		l4_proto_tmp = l3.v4->protocol;
792 	} else {
793 		return -EINVAL;
794 	}
795 
796 	*ol4_proto = l4_proto_tmp;
797 
798 	/* tunnel packet */
799 	if (!skb->encapsulation) {
800 		*il4_proto = 0;
801 		return 0;
802 	}
803 
804 	/* find inner header point */
805 	l3.hdr = skb_inner_network_header(skb);
806 	l4_hdr = skb_inner_transport_header(skb);
807 
808 	if (l3.v6->version == 6) {
809 		exthdr = l3.hdr + sizeof(*l3.v6);
810 		l4_proto_tmp = l3.v6->nexthdr;
811 		if (l4_hdr != exthdr)
812 			ipv6_skip_exthdr(skb, exthdr - skb->data,
813 					 &l4_proto_tmp, &frag_off);
814 	} else if (l3.v4->version == 4) {
815 		l4_proto_tmp = l3.v4->protocol;
816 	}
817 
818 	*il4_proto = l4_proto_tmp;
819 
820 	return 0;
821 }
822 
823 /* when skb->encapsulation is 0, skb->ip_summed is CHECKSUM_PARTIAL
824  * and it is udp packet, which has a dest port as the IANA assigned.
825  * the hardware is expected to do the checksum offload, but the
826  * hardware will not do the checksum offload when udp dest port is
827  * 4789 or 6081.
828  */
829 static bool hns3_tunnel_csum_bug(struct sk_buff *skb)
830 {
831 	struct hns3_nic_priv *priv = netdev_priv(skb->dev);
832 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(priv->ae_handle->pdev);
833 	union l4_hdr_info l4;
834 
835 	/* device version above V3(include V3), the hardware can
836 	 * do this checksum offload.
837 	 */
838 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
839 		return false;
840 
841 	l4.hdr = skb_transport_header(skb);
842 
843 	if (!(!skb->encapsulation &&
844 	      (l4.udp->dest == htons(IANA_VXLAN_UDP_PORT) ||
845 	      l4.udp->dest == htons(GENEVE_UDP_PORT))))
846 		return false;
847 
848 	skb_checksum_help(skb);
849 
850 	return true;
851 }
852 
853 static void hns3_set_outer_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
854 				  u32 *ol_type_vlan_len_msec)
855 {
856 	u32 l2_len, l3_len, l4_len;
857 	unsigned char *il2_hdr;
858 	union l3_hdr_info l3;
859 	union l4_hdr_info l4;
860 
861 	l3.hdr = skb_network_header(skb);
862 	l4.hdr = skb_transport_header(skb);
863 
864 	/* compute OL2 header size, defined in 2 Bytes */
865 	l2_len = l3.hdr - skb->data;
866 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L2LEN_S, l2_len >> 1);
867 
868 	/* compute OL3 header size, defined in 4 Bytes */
869 	l3_len = l4.hdr - l3.hdr;
870 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L3LEN_S, l3_len >> 2);
871 
872 	il2_hdr = skb_inner_mac_header(skb);
873 	/* compute OL4 header size, defined in 4 Bytes */
874 	l4_len = il2_hdr - l4.hdr;
875 	hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_L4LEN_S, l4_len >> 2);
876 
877 	/* define outer network header type */
878 	if (skb->protocol == htons(ETH_P_IP)) {
879 		if (skb_is_gso(skb))
880 			hns3_set_field(*ol_type_vlan_len_msec,
881 				       HNS3_TXD_OL3T_S,
882 				       HNS3_OL3T_IPV4_CSUM);
883 		else
884 			hns3_set_field(*ol_type_vlan_len_msec,
885 				       HNS3_TXD_OL3T_S,
886 				       HNS3_OL3T_IPV4_NO_CSUM);
887 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
888 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_OL3T_S,
889 			       HNS3_OL3T_IPV6);
890 	}
891 
892 	if (ol4_proto == IPPROTO_UDP)
893 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
894 			       HNS3_TUN_MAC_IN_UDP);
895 	else if (ol4_proto == IPPROTO_GRE)
896 		hns3_set_field(*ol_type_vlan_len_msec, HNS3_TXD_TUNTYPE_S,
897 			       HNS3_TUN_NVGRE);
898 }
899 
900 static int hns3_set_l2l3l4(struct sk_buff *skb, u8 ol4_proto,
901 			   u8 il4_proto, u32 *type_cs_vlan_tso,
902 			   u32 *ol_type_vlan_len_msec)
903 {
904 	unsigned char *l2_hdr = skb->data;
905 	u32 l4_proto = ol4_proto;
906 	union l4_hdr_info l4;
907 	union l3_hdr_info l3;
908 	u32 l2_len, l3_len;
909 
910 	l4.hdr = skb_transport_header(skb);
911 	l3.hdr = skb_network_header(skb);
912 
913 	/* handle encapsulation skb */
914 	if (skb->encapsulation) {
915 		/* If this is a not UDP/GRE encapsulation skb */
916 		if (!(ol4_proto == IPPROTO_UDP || ol4_proto == IPPROTO_GRE)) {
917 			/* drop the skb tunnel packet if hardware don't support,
918 			 * because hardware can't calculate csum when TSO.
919 			 */
920 			if (skb_is_gso(skb))
921 				return -EDOM;
922 
923 			/* the stack computes the IP header already,
924 			 * driver calculate l4 checksum when not TSO.
925 			 */
926 			skb_checksum_help(skb);
927 			return 0;
928 		}
929 
930 		hns3_set_outer_l2l3l4(skb, ol4_proto, ol_type_vlan_len_msec);
931 
932 		/* switch to inner header */
933 		l2_hdr = skb_inner_mac_header(skb);
934 		l3.hdr = skb_inner_network_header(skb);
935 		l4.hdr = skb_inner_transport_header(skb);
936 		l4_proto = il4_proto;
937 	}
938 
939 	if (l3.v4->version == 4) {
940 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
941 			       HNS3_L3T_IPV4);
942 
943 		/* the stack computes the IP header already, the only time we
944 		 * need the hardware to recompute it is in the case of TSO.
945 		 */
946 		if (skb_is_gso(skb))
947 			hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3CS_B, 1);
948 	} else if (l3.v6->version == 6) {
949 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3T_S,
950 			       HNS3_L3T_IPV6);
951 	}
952 
953 	/* compute inner(/normal) L2 header size, defined in 2 Bytes */
954 	l2_len = l3.hdr - l2_hdr;
955 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L2LEN_S, l2_len >> 1);
956 
957 	/* compute inner(/normal) L3 header size, defined in 4 Bytes */
958 	l3_len = l4.hdr - l3.hdr;
959 	hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L3LEN_S, l3_len >> 2);
960 
961 	/* compute inner(/normal) L4 header size, defined in 4 Bytes */
962 	switch (l4_proto) {
963 	case IPPROTO_TCP:
964 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
965 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
966 			       HNS3_L4T_TCP);
967 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
968 			       l4.tcp->doff);
969 		break;
970 	case IPPROTO_UDP:
971 		if (hns3_tunnel_csum_bug(skb))
972 			break;
973 
974 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
975 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
976 			       HNS3_L4T_UDP);
977 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
978 			       (sizeof(struct udphdr) >> 2));
979 		break;
980 	case IPPROTO_SCTP:
981 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4CS_B, 1);
982 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4T_S,
983 			       HNS3_L4T_SCTP);
984 		hns3_set_field(*type_cs_vlan_tso, HNS3_TXD_L4LEN_S,
985 			       (sizeof(struct sctphdr) >> 2));
986 		break;
987 	default:
988 		/* drop the skb tunnel packet if hardware don't support,
989 		 * because hardware can't calculate csum when TSO.
990 		 */
991 		if (skb_is_gso(skb))
992 			return -EDOM;
993 
994 		/* the stack computes the IP header already,
995 		 * driver calculate l4 checksum when not TSO.
996 		 */
997 		skb_checksum_help(skb);
998 		return 0;
999 	}
1000 
1001 	return 0;
1002 }
1003 
1004 static int hns3_handle_vtags(struct hns3_enet_ring *tx_ring,
1005 			     struct sk_buff *skb)
1006 {
1007 	struct hnae3_handle *handle = tx_ring->tqp->handle;
1008 	struct hnae3_ae_dev *ae_dev;
1009 	struct vlan_ethhdr *vhdr;
1010 	int rc;
1011 
1012 	if (!(skb->protocol == htons(ETH_P_8021Q) ||
1013 	      skb_vlan_tag_present(skb)))
1014 		return 0;
1015 
1016 	/* For HW limitation on HNAE3_DEVICE_VERSION_V2, if port based insert
1017 	 * VLAN enabled, only one VLAN header is allowed in skb, otherwise it
1018 	 * will cause RAS error.
1019 	 */
1020 	ae_dev = pci_get_drvdata(handle->pdev);
1021 	if (unlikely(skb_vlan_tagged_multi(skb) &&
1022 		     ae_dev->dev_version <= HNAE3_DEVICE_VERSION_V2 &&
1023 		     handle->port_base_vlan_state ==
1024 		     HNAE3_PORT_BASE_VLAN_ENABLE))
1025 		return -EINVAL;
1026 
1027 	if (skb->protocol == htons(ETH_P_8021Q) &&
1028 	    !(handle->kinfo.netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1029 		/* When HW VLAN acceleration is turned off, and the stack
1030 		 * sets the protocol to 802.1q, the driver just need to
1031 		 * set the protocol to the encapsulated ethertype.
1032 		 */
1033 		skb->protocol = vlan_get_protocol(skb);
1034 		return 0;
1035 	}
1036 
1037 	if (skb_vlan_tag_present(skb)) {
1038 		/* Based on hw strategy, use out_vtag in two layer tag case,
1039 		 * and use inner_vtag in one tag case.
1040 		 */
1041 		if (skb->protocol == htons(ETH_P_8021Q) &&
1042 		    handle->port_base_vlan_state ==
1043 		    HNAE3_PORT_BASE_VLAN_DISABLE)
1044 			rc = HNS3_OUTER_VLAN_TAG;
1045 		else
1046 			rc = HNS3_INNER_VLAN_TAG;
1047 
1048 		skb->protocol = vlan_get_protocol(skb);
1049 		return rc;
1050 	}
1051 
1052 	rc = skb_cow_head(skb, 0);
1053 	if (unlikely(rc < 0))
1054 		return rc;
1055 
1056 	vhdr = (struct vlan_ethhdr *)skb->data;
1057 	vhdr->h_vlan_TCI |= cpu_to_be16((skb->priority << VLAN_PRIO_SHIFT)
1058 					 & VLAN_PRIO_MASK);
1059 
1060 	skb->protocol = vlan_get_protocol(skb);
1061 	return 0;
1062 }
1063 
1064 /* check if the hardware is capable of checksum offloading */
1065 static bool hns3_check_hw_tx_csum(struct sk_buff *skb)
1066 {
1067 	struct hns3_nic_priv *priv = netdev_priv(skb->dev);
1068 
1069 	/* Kindly note, due to backward compatibility of the TX descriptor,
1070 	 * HW checksum of the non-IP packets and GSO packets is handled at
1071 	 * different place in the following code
1072 	 */
1073 	if (skb_csum_is_sctp(skb) || skb_is_gso(skb) ||
1074 	    !test_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state))
1075 		return false;
1076 
1077 	return true;
1078 }
1079 
1080 static int hns3_fill_skb_desc(struct hns3_enet_ring *ring,
1081 			      struct sk_buff *skb, struct hns3_desc *desc,
1082 			      struct hns3_desc_cb *desc_cb)
1083 {
1084 	u32 ol_type_vlan_len_msec = 0;
1085 	u32 paylen_ol4cs = skb->len;
1086 	u32 type_cs_vlan_tso = 0;
1087 	u16 mss_hw_csum = 0;
1088 	u16 inner_vtag = 0;
1089 	u16 out_vtag = 0;
1090 	int ret;
1091 
1092 	ret = hns3_handle_vtags(ring, skb);
1093 	if (unlikely(ret < 0)) {
1094 		u64_stats_update_begin(&ring->syncp);
1095 		ring->stats.tx_vlan_err++;
1096 		u64_stats_update_end(&ring->syncp);
1097 		return ret;
1098 	} else if (ret == HNS3_INNER_VLAN_TAG) {
1099 		inner_vtag = skb_vlan_tag_get(skb);
1100 		inner_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1101 				VLAN_PRIO_MASK;
1102 		hns3_set_field(type_cs_vlan_tso, HNS3_TXD_VLAN_B, 1);
1103 	} else if (ret == HNS3_OUTER_VLAN_TAG) {
1104 		out_vtag = skb_vlan_tag_get(skb);
1105 		out_vtag |= (skb->priority << VLAN_PRIO_SHIFT) &
1106 				VLAN_PRIO_MASK;
1107 		hns3_set_field(ol_type_vlan_len_msec, HNS3_TXD_OVLAN_B,
1108 			       1);
1109 	}
1110 
1111 	desc_cb->send_bytes = skb->len;
1112 
1113 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1114 		u8 ol4_proto, il4_proto;
1115 
1116 		if (hns3_check_hw_tx_csum(skb)) {
1117 			/* set checksum start and offset, defined in 2 Bytes */
1118 			hns3_set_field(type_cs_vlan_tso, HNS3_TXD_CSUM_START_S,
1119 				       skb_checksum_start_offset(skb) >> 1);
1120 			hns3_set_field(ol_type_vlan_len_msec,
1121 				       HNS3_TXD_CSUM_OFFSET_S,
1122 				       skb->csum_offset >> 1);
1123 			mss_hw_csum |= BIT(HNS3_TXD_HW_CS_B);
1124 			goto out_hw_tx_csum;
1125 		}
1126 
1127 		skb_reset_mac_len(skb);
1128 
1129 		ret = hns3_get_l4_protocol(skb, &ol4_proto, &il4_proto);
1130 		if (unlikely(ret < 0)) {
1131 			u64_stats_update_begin(&ring->syncp);
1132 			ring->stats.tx_l4_proto_err++;
1133 			u64_stats_update_end(&ring->syncp);
1134 			return ret;
1135 		}
1136 
1137 		ret = hns3_set_l2l3l4(skb, ol4_proto, il4_proto,
1138 				      &type_cs_vlan_tso,
1139 				      &ol_type_vlan_len_msec);
1140 		if (unlikely(ret < 0)) {
1141 			u64_stats_update_begin(&ring->syncp);
1142 			ring->stats.tx_l2l3l4_err++;
1143 			u64_stats_update_end(&ring->syncp);
1144 			return ret;
1145 		}
1146 
1147 		ret = hns3_set_tso(skb, &paylen_ol4cs, &mss_hw_csum,
1148 				   &type_cs_vlan_tso, &desc_cb->send_bytes);
1149 		if (unlikely(ret < 0)) {
1150 			u64_stats_update_begin(&ring->syncp);
1151 			ring->stats.tx_tso_err++;
1152 			u64_stats_update_end(&ring->syncp);
1153 			return ret;
1154 		}
1155 	}
1156 
1157 out_hw_tx_csum:
1158 	/* Set txbd */
1159 	desc->tx.ol_type_vlan_len_msec =
1160 		cpu_to_le32(ol_type_vlan_len_msec);
1161 	desc->tx.type_cs_vlan_tso_len = cpu_to_le32(type_cs_vlan_tso);
1162 	desc->tx.paylen_ol4cs = cpu_to_le32(paylen_ol4cs);
1163 	desc->tx.mss_hw_csum = cpu_to_le16(mss_hw_csum);
1164 	desc->tx.vlan_tag = cpu_to_le16(inner_vtag);
1165 	desc->tx.outer_vlan_tag = cpu_to_le16(out_vtag);
1166 
1167 	return 0;
1168 }
1169 
1170 static int hns3_fill_desc(struct hns3_enet_ring *ring, void *priv,
1171 			  unsigned int size, enum hns_desc_type type)
1172 {
1173 #define HNS3_LIKELY_BD_NUM	1
1174 
1175 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1176 	struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1177 	struct device *dev = ring_to_dev(ring);
1178 	skb_frag_t *frag;
1179 	unsigned int frag_buf_num;
1180 	int k, sizeoflast;
1181 	dma_addr_t dma;
1182 
1183 	if (type == DESC_TYPE_FRAGLIST_SKB ||
1184 	    type == DESC_TYPE_SKB) {
1185 		struct sk_buff *skb = (struct sk_buff *)priv;
1186 
1187 		dma = dma_map_single(dev, skb->data, size, DMA_TO_DEVICE);
1188 	} else {
1189 		frag = (skb_frag_t *)priv;
1190 		dma = skb_frag_dma_map(dev, frag, 0, size, DMA_TO_DEVICE);
1191 	}
1192 
1193 	if (unlikely(dma_mapping_error(dev, dma))) {
1194 		u64_stats_update_begin(&ring->syncp);
1195 		ring->stats.sw_err_cnt++;
1196 		u64_stats_update_end(&ring->syncp);
1197 		return -ENOMEM;
1198 	}
1199 
1200 	desc_cb->priv = priv;
1201 	desc_cb->length = size;
1202 	desc_cb->dma = dma;
1203 	desc_cb->type = type;
1204 
1205 	if (likely(size <= HNS3_MAX_BD_SIZE)) {
1206 		desc->addr = cpu_to_le64(dma);
1207 		desc->tx.send_size = cpu_to_le16(size);
1208 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1209 			cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1210 
1211 		trace_hns3_tx_desc(ring, ring->next_to_use);
1212 		ring_ptr_move_fw(ring, next_to_use);
1213 		return HNS3_LIKELY_BD_NUM;
1214 	}
1215 
1216 	frag_buf_num = hns3_tx_bd_count(size);
1217 	sizeoflast = size % HNS3_MAX_BD_SIZE;
1218 	sizeoflast = sizeoflast ? sizeoflast : HNS3_MAX_BD_SIZE;
1219 
1220 	/* When frag size is bigger than hardware limit, split this frag */
1221 	for (k = 0; k < frag_buf_num; k++) {
1222 		/* now, fill the descriptor */
1223 		desc->addr = cpu_to_le64(dma + HNS3_MAX_BD_SIZE * k);
1224 		desc->tx.send_size = cpu_to_le16((k == frag_buf_num - 1) ?
1225 				     (u16)sizeoflast : (u16)HNS3_MAX_BD_SIZE);
1226 		desc->tx.bdtp_fe_sc_vld_ra_ri =
1227 				cpu_to_le16(BIT(HNS3_TXD_VLD_B));
1228 
1229 		trace_hns3_tx_desc(ring, ring->next_to_use);
1230 		/* move ring pointer to next */
1231 		ring_ptr_move_fw(ring, next_to_use);
1232 
1233 		desc = &ring->desc[ring->next_to_use];
1234 	}
1235 
1236 	return frag_buf_num;
1237 }
1238 
1239 static unsigned int hns3_skb_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1240 				    unsigned int bd_num)
1241 {
1242 	unsigned int size;
1243 	int i;
1244 
1245 	size = skb_headlen(skb);
1246 	while (size > HNS3_MAX_BD_SIZE) {
1247 		bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1248 		size -= HNS3_MAX_BD_SIZE;
1249 
1250 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1251 			return bd_num;
1252 	}
1253 
1254 	if (size) {
1255 		bd_size[bd_num++] = size;
1256 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1257 			return bd_num;
1258 	}
1259 
1260 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1261 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1262 		size = skb_frag_size(frag);
1263 		if (!size)
1264 			continue;
1265 
1266 		while (size > HNS3_MAX_BD_SIZE) {
1267 			bd_size[bd_num++] = HNS3_MAX_BD_SIZE;
1268 			size -= HNS3_MAX_BD_SIZE;
1269 
1270 			if (bd_num > HNS3_MAX_TSO_BD_NUM)
1271 				return bd_num;
1272 		}
1273 
1274 		bd_size[bd_num++] = size;
1275 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1276 			return bd_num;
1277 	}
1278 
1279 	return bd_num;
1280 }
1281 
1282 static unsigned int hns3_tx_bd_num(struct sk_buff *skb, unsigned int *bd_size,
1283 				   u8 max_non_tso_bd_num, unsigned int bd_num,
1284 				   unsigned int recursion_level)
1285 {
1286 #define HNS3_MAX_RECURSION_LEVEL	24
1287 
1288 	struct sk_buff *frag_skb;
1289 
1290 	/* If the total len is within the max bd limit */
1291 	if (likely(skb->len <= HNS3_MAX_BD_SIZE && !recursion_level &&
1292 		   !skb_has_frag_list(skb) &&
1293 		   skb_shinfo(skb)->nr_frags < max_non_tso_bd_num))
1294 		return skb_shinfo(skb)->nr_frags + 1U;
1295 
1296 	if (unlikely(recursion_level >= HNS3_MAX_RECURSION_LEVEL))
1297 		return UINT_MAX;
1298 
1299 	bd_num = hns3_skb_bd_num(skb, bd_size, bd_num);
1300 	if (!skb_has_frag_list(skb) || bd_num > HNS3_MAX_TSO_BD_NUM)
1301 		return bd_num;
1302 
1303 	skb_walk_frags(skb, frag_skb) {
1304 		bd_num = hns3_tx_bd_num(frag_skb, bd_size, max_non_tso_bd_num,
1305 					bd_num, recursion_level + 1);
1306 		if (bd_num > HNS3_MAX_TSO_BD_NUM)
1307 			return bd_num;
1308 	}
1309 
1310 	return bd_num;
1311 }
1312 
1313 static unsigned int hns3_gso_hdr_len(struct sk_buff *skb)
1314 {
1315 	if (!skb->encapsulation)
1316 		return skb_transport_offset(skb) + tcp_hdrlen(skb);
1317 
1318 	return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
1319 }
1320 
1321 /* HW need every continuous max_non_tso_bd_num buffer data to be larger
1322  * than MSS, we simplify it by ensuring skb_headlen + the first continuous
1323  * max_non_tso_bd_num - 1 frags to be larger than gso header len + mss,
1324  * and the remaining continuous max_non_tso_bd_num - 1 frags to be larger
1325  * than MSS except the last max_non_tso_bd_num - 1 frags.
1326  */
1327 static bool hns3_skb_need_linearized(struct sk_buff *skb, unsigned int *bd_size,
1328 				     unsigned int bd_num, u8 max_non_tso_bd_num)
1329 {
1330 	unsigned int tot_len = 0;
1331 	int i;
1332 
1333 	for (i = 0; i < max_non_tso_bd_num - 1U; i++)
1334 		tot_len += bd_size[i];
1335 
1336 	/* ensure the first max_non_tso_bd_num frags is greater than
1337 	 * mss + header
1338 	 */
1339 	if (tot_len + bd_size[max_non_tso_bd_num - 1U] <
1340 	    skb_shinfo(skb)->gso_size + hns3_gso_hdr_len(skb))
1341 		return true;
1342 
1343 	/* ensure every continuous max_non_tso_bd_num - 1 buffer is greater
1344 	 * than mss except the last one.
1345 	 */
1346 	for (i = 0; i < bd_num - max_non_tso_bd_num; i++) {
1347 		tot_len -= bd_size[i];
1348 		tot_len += bd_size[i + max_non_tso_bd_num - 1U];
1349 
1350 		if (tot_len < skb_shinfo(skb)->gso_size)
1351 			return true;
1352 	}
1353 
1354 	return false;
1355 }
1356 
1357 void hns3_shinfo_pack(struct skb_shared_info *shinfo, __u32 *size)
1358 {
1359 	int i;
1360 
1361 	for (i = 0; i < MAX_SKB_FRAGS; i++)
1362 		size[i] = skb_frag_size(&shinfo->frags[i]);
1363 }
1364 
1365 static int hns3_skb_linearize(struct hns3_enet_ring *ring,
1366 			      struct sk_buff *skb,
1367 			      u8 max_non_tso_bd_num,
1368 			      unsigned int bd_num)
1369 {
1370 	/* 'bd_num == UINT_MAX' means the skb' fraglist has a
1371 	 * recursion level of over HNS3_MAX_RECURSION_LEVEL.
1372 	 */
1373 	if (bd_num == UINT_MAX) {
1374 		u64_stats_update_begin(&ring->syncp);
1375 		ring->stats.over_max_recursion++;
1376 		u64_stats_update_end(&ring->syncp);
1377 		return -ENOMEM;
1378 	}
1379 
1380 	/* The skb->len has exceeded the hw limitation, linearization
1381 	 * will not help.
1382 	 */
1383 	if (skb->len > HNS3_MAX_TSO_SIZE ||
1384 	    (!skb_is_gso(skb) && skb->len >
1385 	     HNS3_MAX_NON_TSO_SIZE(max_non_tso_bd_num))) {
1386 		u64_stats_update_begin(&ring->syncp);
1387 		ring->stats.hw_limitation++;
1388 		u64_stats_update_end(&ring->syncp);
1389 		return -ENOMEM;
1390 	}
1391 
1392 	if (__skb_linearize(skb)) {
1393 		u64_stats_update_begin(&ring->syncp);
1394 		ring->stats.sw_err_cnt++;
1395 		u64_stats_update_end(&ring->syncp);
1396 		return -ENOMEM;
1397 	}
1398 
1399 	return 0;
1400 }
1401 
1402 static int hns3_nic_maybe_stop_tx(struct hns3_enet_ring *ring,
1403 				  struct net_device *netdev,
1404 				  struct sk_buff *skb)
1405 {
1406 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1407 	u8 max_non_tso_bd_num = priv->max_non_tso_bd_num;
1408 	unsigned int bd_size[HNS3_MAX_TSO_BD_NUM + 1U];
1409 	unsigned int bd_num;
1410 
1411 	bd_num = hns3_tx_bd_num(skb, bd_size, max_non_tso_bd_num, 0, 0);
1412 	if (unlikely(bd_num > max_non_tso_bd_num)) {
1413 		if (bd_num <= HNS3_MAX_TSO_BD_NUM && skb_is_gso(skb) &&
1414 		    !hns3_skb_need_linearized(skb, bd_size, bd_num,
1415 					      max_non_tso_bd_num)) {
1416 			trace_hns3_over_max_bd(skb);
1417 			goto out;
1418 		}
1419 
1420 		if (hns3_skb_linearize(ring, skb, max_non_tso_bd_num,
1421 				       bd_num))
1422 			return -ENOMEM;
1423 
1424 		bd_num = hns3_tx_bd_count(skb->len);
1425 
1426 		u64_stats_update_begin(&ring->syncp);
1427 		ring->stats.tx_copy++;
1428 		u64_stats_update_end(&ring->syncp);
1429 	}
1430 
1431 out:
1432 	if (likely(ring_space(ring) >= bd_num))
1433 		return bd_num;
1434 
1435 	netif_stop_subqueue(netdev, ring->queue_index);
1436 	smp_mb(); /* Memory barrier before checking ring_space */
1437 
1438 	/* Start queue in case hns3_clean_tx_ring has just made room
1439 	 * available and has not seen the queue stopped state performed
1440 	 * by netif_stop_subqueue above.
1441 	 */
1442 	if (ring_space(ring) >= bd_num && netif_carrier_ok(netdev) &&
1443 	    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
1444 		netif_start_subqueue(netdev, ring->queue_index);
1445 		return bd_num;
1446 	}
1447 
1448 	u64_stats_update_begin(&ring->syncp);
1449 	ring->stats.tx_busy++;
1450 	u64_stats_update_end(&ring->syncp);
1451 
1452 	return -EBUSY;
1453 }
1454 
1455 static void hns3_clear_desc(struct hns3_enet_ring *ring, int next_to_use_orig)
1456 {
1457 	struct device *dev = ring_to_dev(ring);
1458 	unsigned int i;
1459 
1460 	for (i = 0; i < ring->desc_num; i++) {
1461 		struct hns3_desc *desc = &ring->desc[ring->next_to_use];
1462 
1463 		memset(desc, 0, sizeof(*desc));
1464 
1465 		/* check if this is where we started */
1466 		if (ring->next_to_use == next_to_use_orig)
1467 			break;
1468 
1469 		/* rollback one */
1470 		ring_ptr_move_bw(ring, next_to_use);
1471 
1472 		if (!ring->desc_cb[ring->next_to_use].dma)
1473 			continue;
1474 
1475 		/* unmap the descriptor dma address */
1476 		if (ring->desc_cb[ring->next_to_use].type == DESC_TYPE_SKB ||
1477 		    ring->desc_cb[ring->next_to_use].type ==
1478 		    DESC_TYPE_FRAGLIST_SKB)
1479 			dma_unmap_single(dev,
1480 					 ring->desc_cb[ring->next_to_use].dma,
1481 					ring->desc_cb[ring->next_to_use].length,
1482 					DMA_TO_DEVICE);
1483 		else if (ring->desc_cb[ring->next_to_use].length)
1484 			dma_unmap_page(dev,
1485 				       ring->desc_cb[ring->next_to_use].dma,
1486 				       ring->desc_cb[ring->next_to_use].length,
1487 				       DMA_TO_DEVICE);
1488 
1489 		ring->desc_cb[ring->next_to_use].length = 0;
1490 		ring->desc_cb[ring->next_to_use].dma = 0;
1491 		ring->desc_cb[ring->next_to_use].type = DESC_TYPE_UNKNOWN;
1492 	}
1493 }
1494 
1495 static int hns3_fill_skb_to_desc(struct hns3_enet_ring *ring,
1496 				 struct sk_buff *skb, enum hns_desc_type type)
1497 {
1498 	unsigned int size = skb_headlen(skb);
1499 	struct sk_buff *frag_skb;
1500 	int i, ret, bd_num = 0;
1501 
1502 	if (size) {
1503 		ret = hns3_fill_desc(ring, skb, size, type);
1504 		if (unlikely(ret < 0))
1505 			return ret;
1506 
1507 		bd_num += ret;
1508 	}
1509 
1510 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1511 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1512 
1513 		size = skb_frag_size(frag);
1514 		if (!size)
1515 			continue;
1516 
1517 		ret = hns3_fill_desc(ring, frag, size, DESC_TYPE_PAGE);
1518 		if (unlikely(ret < 0))
1519 			return ret;
1520 
1521 		bd_num += ret;
1522 	}
1523 
1524 	skb_walk_frags(skb, frag_skb) {
1525 		ret = hns3_fill_skb_to_desc(ring, frag_skb,
1526 					    DESC_TYPE_FRAGLIST_SKB);
1527 		if (unlikely(ret < 0))
1528 			return ret;
1529 
1530 		bd_num += ret;
1531 	}
1532 
1533 	return bd_num;
1534 }
1535 
1536 static void hns3_tx_doorbell(struct hns3_enet_ring *ring, int num,
1537 			     bool doorbell)
1538 {
1539 	ring->pending_buf += num;
1540 
1541 	if (!doorbell) {
1542 		u64_stats_update_begin(&ring->syncp);
1543 		ring->stats.tx_more++;
1544 		u64_stats_update_end(&ring->syncp);
1545 		return;
1546 	}
1547 
1548 	if (!ring->pending_buf)
1549 		return;
1550 
1551 	writel(ring->pending_buf,
1552 	       ring->tqp->io_base + HNS3_RING_TX_RING_TAIL_REG);
1553 	ring->pending_buf = 0;
1554 	WRITE_ONCE(ring->last_to_use, ring->next_to_use);
1555 }
1556 
1557 netdev_tx_t hns3_nic_net_xmit(struct sk_buff *skb, struct net_device *netdev)
1558 {
1559 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1560 	struct hns3_enet_ring *ring = &priv->ring[skb->queue_mapping];
1561 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_use];
1562 	struct netdev_queue *dev_queue;
1563 	int pre_ntu, next_to_use_head;
1564 	bool doorbell;
1565 	int ret;
1566 
1567 	/* Hardware can only handle short frames above 32 bytes */
1568 	if (skb_put_padto(skb, HNS3_MIN_TX_LEN)) {
1569 		hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1570 
1571 		u64_stats_update_begin(&ring->syncp);
1572 		ring->stats.sw_err_cnt++;
1573 		u64_stats_update_end(&ring->syncp);
1574 
1575 		return NETDEV_TX_OK;
1576 	}
1577 
1578 	/* Prefetch the data used later */
1579 	prefetch(skb->data);
1580 
1581 	ret = hns3_nic_maybe_stop_tx(ring, netdev, skb);
1582 	if (unlikely(ret <= 0)) {
1583 		if (ret == -EBUSY) {
1584 			hns3_tx_doorbell(ring, 0, true);
1585 			return NETDEV_TX_BUSY;
1586 		}
1587 
1588 		hns3_rl_err(netdev, "xmit error: %d!\n", ret);
1589 		goto out_err_tx_ok;
1590 	}
1591 
1592 	next_to_use_head = ring->next_to_use;
1593 
1594 	ret = hns3_fill_skb_desc(ring, skb, &ring->desc[ring->next_to_use],
1595 				 desc_cb);
1596 	if (unlikely(ret < 0))
1597 		goto fill_err;
1598 
1599 	/* 'ret < 0' means filling error, 'ret == 0' means skb->len is
1600 	 * zero, which is unlikely, and 'ret > 0' means how many tx desc
1601 	 * need to be notified to the hw.
1602 	 */
1603 	ret = hns3_fill_skb_to_desc(ring, skb, DESC_TYPE_SKB);
1604 	if (unlikely(ret <= 0))
1605 		goto fill_err;
1606 
1607 	pre_ntu = ring->next_to_use ? (ring->next_to_use - 1) :
1608 					(ring->desc_num - 1);
1609 	ring->desc[pre_ntu].tx.bdtp_fe_sc_vld_ra_ri |=
1610 				cpu_to_le16(BIT(HNS3_TXD_FE_B));
1611 	trace_hns3_tx_desc(ring, pre_ntu);
1612 
1613 	/* Complete translate all packets */
1614 	dev_queue = netdev_get_tx_queue(netdev, ring->queue_index);
1615 	doorbell = __netdev_tx_sent_queue(dev_queue, desc_cb->send_bytes,
1616 					  netdev_xmit_more());
1617 	hns3_tx_doorbell(ring, ret, doorbell);
1618 
1619 	return NETDEV_TX_OK;
1620 
1621 fill_err:
1622 	hns3_clear_desc(ring, next_to_use_head);
1623 
1624 out_err_tx_ok:
1625 	dev_kfree_skb_any(skb);
1626 	hns3_tx_doorbell(ring, 0, !netdev_xmit_more());
1627 	return NETDEV_TX_OK;
1628 }
1629 
1630 static int hns3_nic_net_set_mac_address(struct net_device *netdev, void *p)
1631 {
1632 	struct hnae3_handle *h = hns3_get_handle(netdev);
1633 	struct sockaddr *mac_addr = p;
1634 	int ret;
1635 
1636 	if (!mac_addr || !is_valid_ether_addr((const u8 *)mac_addr->sa_data))
1637 		return -EADDRNOTAVAIL;
1638 
1639 	if (ether_addr_equal(netdev->dev_addr, mac_addr->sa_data)) {
1640 		netdev_info(netdev, "already using mac address %pM\n",
1641 			    mac_addr->sa_data);
1642 		return 0;
1643 	}
1644 
1645 	/* For VF device, if there is a perm_addr, then the user will not
1646 	 * be allowed to change the address.
1647 	 */
1648 	if (!hns3_is_phys_func(h->pdev) &&
1649 	    !is_zero_ether_addr(netdev->perm_addr)) {
1650 		netdev_err(netdev, "has permanent MAC %pM, user MAC %pM not allow\n",
1651 			   netdev->perm_addr, mac_addr->sa_data);
1652 		return -EPERM;
1653 	}
1654 
1655 	ret = h->ae_algo->ops->set_mac_addr(h, mac_addr->sa_data, false);
1656 	if (ret) {
1657 		netdev_err(netdev, "set_mac_address fail, ret=%d!\n", ret);
1658 		return ret;
1659 	}
1660 
1661 	ether_addr_copy(netdev->dev_addr, mac_addr->sa_data);
1662 
1663 	return 0;
1664 }
1665 
1666 static int hns3_nic_do_ioctl(struct net_device *netdev,
1667 			     struct ifreq *ifr, int cmd)
1668 {
1669 	struct hnae3_handle *h = hns3_get_handle(netdev);
1670 
1671 	if (!netif_running(netdev))
1672 		return -EINVAL;
1673 
1674 	if (!h->ae_algo->ops->do_ioctl)
1675 		return -EOPNOTSUPP;
1676 
1677 	return h->ae_algo->ops->do_ioctl(h, ifr, cmd);
1678 }
1679 
1680 static int hns3_nic_set_features(struct net_device *netdev,
1681 				 netdev_features_t features)
1682 {
1683 	netdev_features_t changed = netdev->features ^ features;
1684 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1685 	struct hnae3_handle *h = priv->ae_handle;
1686 	bool enable;
1687 	int ret;
1688 
1689 	if (changed & (NETIF_F_GRO_HW) && h->ae_algo->ops->set_gro_en) {
1690 		enable = !!(features & NETIF_F_GRO_HW);
1691 		ret = h->ae_algo->ops->set_gro_en(h, enable);
1692 		if (ret)
1693 			return ret;
1694 	}
1695 
1696 	if ((changed & NETIF_F_HW_VLAN_CTAG_RX) &&
1697 	    h->ae_algo->ops->enable_hw_strip_rxvtag) {
1698 		enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1699 		ret = h->ae_algo->ops->enable_hw_strip_rxvtag(h, enable);
1700 		if (ret)
1701 			return ret;
1702 	}
1703 
1704 	if ((changed & NETIF_F_NTUPLE) && h->ae_algo->ops->enable_fd) {
1705 		enable = !!(features & NETIF_F_NTUPLE);
1706 		h->ae_algo->ops->enable_fd(h, enable);
1707 	}
1708 
1709 	if ((netdev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC) &&
1710 	    h->ae_algo->ops->cls_flower_active(h)) {
1711 		netdev_err(netdev,
1712 			   "there are offloaded TC filters active, cannot disable HW TC offload");
1713 		return -EINVAL;
1714 	}
1715 
1716 	netdev->features = features;
1717 	return 0;
1718 }
1719 
1720 static netdev_features_t hns3_features_check(struct sk_buff *skb,
1721 					     struct net_device *dev,
1722 					     netdev_features_t features)
1723 {
1724 #define HNS3_MAX_HDR_LEN	480U
1725 #define HNS3_MAX_L4_HDR_LEN	60U
1726 
1727 	size_t len;
1728 
1729 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1730 		return features;
1731 
1732 	if (skb->encapsulation)
1733 		len = skb_inner_transport_header(skb) - skb->data;
1734 	else
1735 		len = skb_transport_header(skb) - skb->data;
1736 
1737 	/* Assume L4 is 60 byte as TCP is the only protocol with a
1738 	 * a flexible value, and it's max len is 60 bytes.
1739 	 */
1740 	len += HNS3_MAX_L4_HDR_LEN;
1741 
1742 	/* Hardware only supports checksum on the skb with a max header
1743 	 * len of 480 bytes.
1744 	 */
1745 	if (len > HNS3_MAX_HDR_LEN)
1746 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
1747 
1748 	return features;
1749 }
1750 
1751 static void hns3_nic_get_stats64(struct net_device *netdev,
1752 				 struct rtnl_link_stats64 *stats)
1753 {
1754 	struct hns3_nic_priv *priv = netdev_priv(netdev);
1755 	int queue_num = priv->ae_handle->kinfo.num_tqps;
1756 	struct hnae3_handle *handle = priv->ae_handle;
1757 	struct hns3_enet_ring *ring;
1758 	u64 rx_length_errors = 0;
1759 	u64 rx_crc_errors = 0;
1760 	u64 rx_multicast = 0;
1761 	unsigned int start;
1762 	u64 tx_errors = 0;
1763 	u64 rx_errors = 0;
1764 	unsigned int idx;
1765 	u64 tx_bytes = 0;
1766 	u64 rx_bytes = 0;
1767 	u64 tx_pkts = 0;
1768 	u64 rx_pkts = 0;
1769 	u64 tx_drop = 0;
1770 	u64 rx_drop = 0;
1771 
1772 	if (test_bit(HNS3_NIC_STATE_DOWN, &priv->state))
1773 		return;
1774 
1775 	handle->ae_algo->ops->update_stats(handle, &netdev->stats);
1776 
1777 	for (idx = 0; idx < queue_num; idx++) {
1778 		/* fetch the tx stats */
1779 		ring = &priv->ring[idx];
1780 		do {
1781 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1782 			tx_bytes += ring->stats.tx_bytes;
1783 			tx_pkts += ring->stats.tx_pkts;
1784 			tx_drop += ring->stats.sw_err_cnt;
1785 			tx_drop += ring->stats.tx_vlan_err;
1786 			tx_drop += ring->stats.tx_l4_proto_err;
1787 			tx_drop += ring->stats.tx_l2l3l4_err;
1788 			tx_drop += ring->stats.tx_tso_err;
1789 			tx_drop += ring->stats.over_max_recursion;
1790 			tx_drop += ring->stats.hw_limitation;
1791 			tx_errors += ring->stats.sw_err_cnt;
1792 			tx_errors += ring->stats.tx_vlan_err;
1793 			tx_errors += ring->stats.tx_l4_proto_err;
1794 			tx_errors += ring->stats.tx_l2l3l4_err;
1795 			tx_errors += ring->stats.tx_tso_err;
1796 			tx_errors += ring->stats.over_max_recursion;
1797 			tx_errors += ring->stats.hw_limitation;
1798 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1799 
1800 		/* fetch the rx stats */
1801 		ring = &priv->ring[idx + queue_num];
1802 		do {
1803 			start = u64_stats_fetch_begin_irq(&ring->syncp);
1804 			rx_bytes += ring->stats.rx_bytes;
1805 			rx_pkts += ring->stats.rx_pkts;
1806 			rx_drop += ring->stats.l2_err;
1807 			rx_errors += ring->stats.l2_err;
1808 			rx_errors += ring->stats.l3l4_csum_err;
1809 			rx_crc_errors += ring->stats.l2_err;
1810 			rx_multicast += ring->stats.rx_multicast;
1811 			rx_length_errors += ring->stats.err_pkt_len;
1812 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
1813 	}
1814 
1815 	stats->tx_bytes = tx_bytes;
1816 	stats->tx_packets = tx_pkts;
1817 	stats->rx_bytes = rx_bytes;
1818 	stats->rx_packets = rx_pkts;
1819 
1820 	stats->rx_errors = rx_errors;
1821 	stats->multicast = rx_multicast;
1822 	stats->rx_length_errors = rx_length_errors;
1823 	stats->rx_crc_errors = rx_crc_errors;
1824 	stats->rx_missed_errors = netdev->stats.rx_missed_errors;
1825 
1826 	stats->tx_errors = tx_errors;
1827 	stats->rx_dropped = rx_drop;
1828 	stats->tx_dropped = tx_drop;
1829 	stats->collisions = netdev->stats.collisions;
1830 	stats->rx_over_errors = netdev->stats.rx_over_errors;
1831 	stats->rx_frame_errors = netdev->stats.rx_frame_errors;
1832 	stats->rx_fifo_errors = netdev->stats.rx_fifo_errors;
1833 	stats->tx_aborted_errors = netdev->stats.tx_aborted_errors;
1834 	stats->tx_carrier_errors = netdev->stats.tx_carrier_errors;
1835 	stats->tx_fifo_errors = netdev->stats.tx_fifo_errors;
1836 	stats->tx_heartbeat_errors = netdev->stats.tx_heartbeat_errors;
1837 	stats->tx_window_errors = netdev->stats.tx_window_errors;
1838 	stats->rx_compressed = netdev->stats.rx_compressed;
1839 	stats->tx_compressed = netdev->stats.tx_compressed;
1840 }
1841 
1842 static int hns3_setup_tc(struct net_device *netdev, void *type_data)
1843 {
1844 	struct tc_mqprio_qopt_offload *mqprio_qopt = type_data;
1845 	struct hnae3_knic_private_info *kinfo;
1846 	u8 tc = mqprio_qopt->qopt.num_tc;
1847 	u16 mode = mqprio_qopt->mode;
1848 	u8 hw = mqprio_qopt->qopt.hw;
1849 	struct hnae3_handle *h;
1850 
1851 	if (!((hw == TC_MQPRIO_HW_OFFLOAD_TCS &&
1852 	       mode == TC_MQPRIO_MODE_CHANNEL) || (!hw && tc == 0)))
1853 		return -EOPNOTSUPP;
1854 
1855 	if (tc > HNAE3_MAX_TC)
1856 		return -EINVAL;
1857 
1858 	if (!netdev)
1859 		return -EINVAL;
1860 
1861 	h = hns3_get_handle(netdev);
1862 	kinfo = &h->kinfo;
1863 
1864 	netif_dbg(h, drv, netdev, "setup tc: num_tc=%u\n", tc);
1865 
1866 	return (kinfo->dcb_ops && kinfo->dcb_ops->setup_tc) ?
1867 		kinfo->dcb_ops->setup_tc(h, mqprio_qopt) : -EOPNOTSUPP;
1868 }
1869 
1870 static int hns3_setup_tc_cls_flower(struct hns3_nic_priv *priv,
1871 				    struct flow_cls_offload *flow)
1872 {
1873 	int tc = tc_classid_to_hwtc(priv->netdev, flow->classid);
1874 	struct hnae3_handle *h = hns3_get_handle(priv->netdev);
1875 
1876 	switch (flow->command) {
1877 	case FLOW_CLS_REPLACE:
1878 		if (h->ae_algo->ops->add_cls_flower)
1879 			return h->ae_algo->ops->add_cls_flower(h, flow, tc);
1880 		break;
1881 	case FLOW_CLS_DESTROY:
1882 		if (h->ae_algo->ops->del_cls_flower)
1883 			return h->ae_algo->ops->del_cls_flower(h, flow);
1884 		break;
1885 	default:
1886 		break;
1887 	}
1888 
1889 	return -EOPNOTSUPP;
1890 }
1891 
1892 static int hns3_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
1893 				  void *cb_priv)
1894 {
1895 	struct hns3_nic_priv *priv = cb_priv;
1896 
1897 	if (!tc_cls_can_offload_and_chain0(priv->netdev, type_data))
1898 		return -EOPNOTSUPP;
1899 
1900 	switch (type) {
1901 	case TC_SETUP_CLSFLOWER:
1902 		return hns3_setup_tc_cls_flower(priv, type_data);
1903 	default:
1904 		return -EOPNOTSUPP;
1905 	}
1906 }
1907 
1908 static LIST_HEAD(hns3_block_cb_list);
1909 
1910 static int hns3_nic_setup_tc(struct net_device *dev, enum tc_setup_type type,
1911 			     void *type_data)
1912 {
1913 	struct hns3_nic_priv *priv = netdev_priv(dev);
1914 	int ret;
1915 
1916 	switch (type) {
1917 	case TC_SETUP_QDISC_MQPRIO:
1918 		ret = hns3_setup_tc(dev, type_data);
1919 		break;
1920 	case TC_SETUP_BLOCK:
1921 		ret = flow_block_cb_setup_simple(type_data,
1922 						 &hns3_block_cb_list,
1923 						 hns3_setup_tc_block_cb,
1924 						 priv, priv, true);
1925 		break;
1926 	default:
1927 		return -EOPNOTSUPP;
1928 	}
1929 
1930 	return ret;
1931 }
1932 
1933 static int hns3_vlan_rx_add_vid(struct net_device *netdev,
1934 				__be16 proto, u16 vid)
1935 {
1936 	struct hnae3_handle *h = hns3_get_handle(netdev);
1937 	int ret = -EIO;
1938 
1939 	if (h->ae_algo->ops->set_vlan_filter)
1940 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, false);
1941 
1942 	return ret;
1943 }
1944 
1945 static int hns3_vlan_rx_kill_vid(struct net_device *netdev,
1946 				 __be16 proto, u16 vid)
1947 {
1948 	struct hnae3_handle *h = hns3_get_handle(netdev);
1949 	int ret = -EIO;
1950 
1951 	if (h->ae_algo->ops->set_vlan_filter)
1952 		ret = h->ae_algo->ops->set_vlan_filter(h, proto, vid, true);
1953 
1954 	return ret;
1955 }
1956 
1957 static int hns3_ndo_set_vf_vlan(struct net_device *netdev, int vf, u16 vlan,
1958 				u8 qos, __be16 vlan_proto)
1959 {
1960 	struct hnae3_handle *h = hns3_get_handle(netdev);
1961 	int ret = -EIO;
1962 
1963 	netif_dbg(h, drv, netdev,
1964 		  "set vf vlan: vf=%d, vlan=%u, qos=%u, vlan_proto=0x%x\n",
1965 		  vf, vlan, qos, ntohs(vlan_proto));
1966 
1967 	if (h->ae_algo->ops->set_vf_vlan_filter)
1968 		ret = h->ae_algo->ops->set_vf_vlan_filter(h, vf, vlan,
1969 							  qos, vlan_proto);
1970 
1971 	return ret;
1972 }
1973 
1974 static int hns3_set_vf_spoofchk(struct net_device *netdev, int vf, bool enable)
1975 {
1976 	struct hnae3_handle *handle = hns3_get_handle(netdev);
1977 
1978 	if (hns3_nic_resetting(netdev))
1979 		return -EBUSY;
1980 
1981 	if (!handle->ae_algo->ops->set_vf_spoofchk)
1982 		return -EOPNOTSUPP;
1983 
1984 	return handle->ae_algo->ops->set_vf_spoofchk(handle, vf, enable);
1985 }
1986 
1987 static int hns3_set_vf_trust(struct net_device *netdev, int vf, bool enable)
1988 {
1989 	struct hnae3_handle *handle = hns3_get_handle(netdev);
1990 
1991 	if (!handle->ae_algo->ops->set_vf_trust)
1992 		return -EOPNOTSUPP;
1993 
1994 	return handle->ae_algo->ops->set_vf_trust(handle, vf, enable);
1995 }
1996 
1997 static int hns3_nic_change_mtu(struct net_device *netdev, int new_mtu)
1998 {
1999 	struct hnae3_handle *h = hns3_get_handle(netdev);
2000 	int ret;
2001 
2002 	if (hns3_nic_resetting(netdev))
2003 		return -EBUSY;
2004 
2005 	if (!h->ae_algo->ops->set_mtu)
2006 		return -EOPNOTSUPP;
2007 
2008 	netif_dbg(h, drv, netdev,
2009 		  "change mtu from %u to %d\n", netdev->mtu, new_mtu);
2010 
2011 	ret = h->ae_algo->ops->set_mtu(h, new_mtu);
2012 	if (ret)
2013 		netdev_err(netdev, "failed to change MTU in hardware %d\n",
2014 			   ret);
2015 	else
2016 		netdev->mtu = new_mtu;
2017 
2018 	return ret;
2019 }
2020 
2021 static bool hns3_get_tx_timeo_queue_info(struct net_device *ndev)
2022 {
2023 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2024 	struct hnae3_handle *h = hns3_get_handle(ndev);
2025 	struct hns3_enet_ring *tx_ring;
2026 	struct napi_struct *napi;
2027 	int timeout_queue = 0;
2028 	int hw_head, hw_tail;
2029 	int fbd_num, fbd_oft;
2030 	int ebd_num, ebd_oft;
2031 	int bd_num, bd_err;
2032 	int ring_en, tc;
2033 	int i;
2034 
2035 	/* Find the stopped queue the same way the stack does */
2036 	for (i = 0; i < ndev->num_tx_queues; i++) {
2037 		struct netdev_queue *q;
2038 		unsigned long trans_start;
2039 
2040 		q = netdev_get_tx_queue(ndev, i);
2041 		trans_start = q->trans_start;
2042 		if (netif_xmit_stopped(q) &&
2043 		    time_after(jiffies,
2044 			       (trans_start + ndev->watchdog_timeo))) {
2045 			timeout_queue = i;
2046 			netdev_info(ndev, "queue state: 0x%lx, delta msecs: %u\n",
2047 				    q->state,
2048 				    jiffies_to_msecs(jiffies - trans_start));
2049 			break;
2050 		}
2051 	}
2052 
2053 	if (i == ndev->num_tx_queues) {
2054 		netdev_info(ndev,
2055 			    "no netdev TX timeout queue found, timeout count: %llu\n",
2056 			    priv->tx_timeout_count);
2057 		return false;
2058 	}
2059 
2060 	priv->tx_timeout_count++;
2061 
2062 	tx_ring = &priv->ring[timeout_queue];
2063 	napi = &tx_ring->tqp_vector->napi;
2064 
2065 	netdev_info(ndev,
2066 		    "tx_timeout count: %llu, queue id: %d, SW_NTU: 0x%x, SW_NTC: 0x%x, napi state: %lu\n",
2067 		    priv->tx_timeout_count, timeout_queue, tx_ring->next_to_use,
2068 		    tx_ring->next_to_clean, napi->state);
2069 
2070 	netdev_info(ndev,
2071 		    "tx_pkts: %llu, tx_bytes: %llu, sw_err_cnt: %llu, tx_pending: %d\n",
2072 		    tx_ring->stats.tx_pkts, tx_ring->stats.tx_bytes,
2073 		    tx_ring->stats.sw_err_cnt, tx_ring->pending_buf);
2074 
2075 	netdev_info(ndev,
2076 		    "seg_pkt_cnt: %llu, tx_more: %llu, restart_queue: %llu, tx_busy: %llu\n",
2077 		    tx_ring->stats.seg_pkt_cnt, tx_ring->stats.tx_more,
2078 		    tx_ring->stats.restart_queue, tx_ring->stats.tx_busy);
2079 
2080 	/* When mac received many pause frames continuous, it's unable to send
2081 	 * packets, which may cause tx timeout
2082 	 */
2083 	if (h->ae_algo->ops->get_mac_stats) {
2084 		struct hns3_mac_stats mac_stats;
2085 
2086 		h->ae_algo->ops->get_mac_stats(h, &mac_stats);
2087 		netdev_info(ndev, "tx_pause_cnt: %llu, rx_pause_cnt: %llu\n",
2088 			    mac_stats.tx_pause_cnt, mac_stats.rx_pause_cnt);
2089 	}
2090 
2091 	hw_head = readl_relaxed(tx_ring->tqp->io_base +
2092 				HNS3_RING_TX_RING_HEAD_REG);
2093 	hw_tail = readl_relaxed(tx_ring->tqp->io_base +
2094 				HNS3_RING_TX_RING_TAIL_REG);
2095 	fbd_num = readl_relaxed(tx_ring->tqp->io_base +
2096 				HNS3_RING_TX_RING_FBDNUM_REG);
2097 	fbd_oft = readl_relaxed(tx_ring->tqp->io_base +
2098 				HNS3_RING_TX_RING_OFFSET_REG);
2099 	ebd_num = readl_relaxed(tx_ring->tqp->io_base +
2100 				HNS3_RING_TX_RING_EBDNUM_REG);
2101 	ebd_oft = readl_relaxed(tx_ring->tqp->io_base +
2102 				HNS3_RING_TX_RING_EBD_OFFSET_REG);
2103 	bd_num = readl_relaxed(tx_ring->tqp->io_base +
2104 			       HNS3_RING_TX_RING_BD_NUM_REG);
2105 	bd_err = readl_relaxed(tx_ring->tqp->io_base +
2106 			       HNS3_RING_TX_RING_BD_ERR_REG);
2107 	ring_en = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_EN_REG);
2108 	tc = readl_relaxed(tx_ring->tqp->io_base + HNS3_RING_TX_RING_TC_REG);
2109 
2110 	netdev_info(ndev,
2111 		    "BD_NUM: 0x%x HW_HEAD: 0x%x, HW_TAIL: 0x%x, BD_ERR: 0x%x, INT: 0x%x\n",
2112 		    bd_num, hw_head, hw_tail, bd_err,
2113 		    readl(tx_ring->tqp_vector->mask_addr));
2114 	netdev_info(ndev,
2115 		    "RING_EN: 0x%x, TC: 0x%x, FBD_NUM: 0x%x FBD_OFT: 0x%x, EBD_NUM: 0x%x, EBD_OFT: 0x%x\n",
2116 		    ring_en, tc, fbd_num, fbd_oft, ebd_num, ebd_oft);
2117 
2118 	return true;
2119 }
2120 
2121 static void hns3_nic_net_timeout(struct net_device *ndev, unsigned int txqueue)
2122 {
2123 	struct hns3_nic_priv *priv = netdev_priv(ndev);
2124 	struct hnae3_handle *h = priv->ae_handle;
2125 
2126 	if (!hns3_get_tx_timeo_queue_info(ndev))
2127 		return;
2128 
2129 	/* request the reset, and let the hclge to determine
2130 	 * which reset level should be done
2131 	 */
2132 	if (h->ae_algo->ops->reset_event)
2133 		h->ae_algo->ops->reset_event(h->pdev, h);
2134 }
2135 
2136 #ifdef CONFIG_RFS_ACCEL
2137 static int hns3_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
2138 			      u16 rxq_index, u32 flow_id)
2139 {
2140 	struct hnae3_handle *h = hns3_get_handle(dev);
2141 	struct flow_keys fkeys;
2142 
2143 	if (!h->ae_algo->ops->add_arfs_entry)
2144 		return -EOPNOTSUPP;
2145 
2146 	if (skb->encapsulation)
2147 		return -EPROTONOSUPPORT;
2148 
2149 	if (!skb_flow_dissect_flow_keys(skb, &fkeys, 0))
2150 		return -EPROTONOSUPPORT;
2151 
2152 	if ((fkeys.basic.n_proto != htons(ETH_P_IP) &&
2153 	     fkeys.basic.n_proto != htons(ETH_P_IPV6)) ||
2154 	    (fkeys.basic.ip_proto != IPPROTO_TCP &&
2155 	     fkeys.basic.ip_proto != IPPROTO_UDP))
2156 		return -EPROTONOSUPPORT;
2157 
2158 	return h->ae_algo->ops->add_arfs_entry(h, rxq_index, flow_id, &fkeys);
2159 }
2160 #endif
2161 
2162 static int hns3_nic_get_vf_config(struct net_device *ndev, int vf,
2163 				  struct ifla_vf_info *ivf)
2164 {
2165 	struct hnae3_handle *h = hns3_get_handle(ndev);
2166 
2167 	if (!h->ae_algo->ops->get_vf_config)
2168 		return -EOPNOTSUPP;
2169 
2170 	return h->ae_algo->ops->get_vf_config(h, vf, ivf);
2171 }
2172 
2173 static int hns3_nic_set_vf_link_state(struct net_device *ndev, int vf,
2174 				      int link_state)
2175 {
2176 	struct hnae3_handle *h = hns3_get_handle(ndev);
2177 
2178 	if (!h->ae_algo->ops->set_vf_link_state)
2179 		return -EOPNOTSUPP;
2180 
2181 	return h->ae_algo->ops->set_vf_link_state(h, vf, link_state);
2182 }
2183 
2184 static int hns3_nic_set_vf_rate(struct net_device *ndev, int vf,
2185 				int min_tx_rate, int max_tx_rate)
2186 {
2187 	struct hnae3_handle *h = hns3_get_handle(ndev);
2188 
2189 	if (!h->ae_algo->ops->set_vf_rate)
2190 		return -EOPNOTSUPP;
2191 
2192 	return h->ae_algo->ops->set_vf_rate(h, vf, min_tx_rate, max_tx_rate,
2193 					    false);
2194 }
2195 
2196 static int hns3_nic_set_vf_mac(struct net_device *netdev, int vf_id, u8 *mac)
2197 {
2198 	struct hnae3_handle *h = hns3_get_handle(netdev);
2199 
2200 	if (!h->ae_algo->ops->set_vf_mac)
2201 		return -EOPNOTSUPP;
2202 
2203 	if (is_multicast_ether_addr(mac)) {
2204 		netdev_err(netdev,
2205 			   "Invalid MAC:%pM specified. Could not set MAC\n",
2206 			   mac);
2207 		return -EINVAL;
2208 	}
2209 
2210 	return h->ae_algo->ops->set_vf_mac(h, vf_id, mac);
2211 }
2212 
2213 static const struct net_device_ops hns3_nic_netdev_ops = {
2214 	.ndo_open		= hns3_nic_net_open,
2215 	.ndo_stop		= hns3_nic_net_stop,
2216 	.ndo_start_xmit		= hns3_nic_net_xmit,
2217 	.ndo_tx_timeout		= hns3_nic_net_timeout,
2218 	.ndo_set_mac_address	= hns3_nic_net_set_mac_address,
2219 	.ndo_do_ioctl		= hns3_nic_do_ioctl,
2220 	.ndo_change_mtu		= hns3_nic_change_mtu,
2221 	.ndo_set_features	= hns3_nic_set_features,
2222 	.ndo_features_check	= hns3_features_check,
2223 	.ndo_get_stats64	= hns3_nic_get_stats64,
2224 	.ndo_setup_tc		= hns3_nic_setup_tc,
2225 	.ndo_set_rx_mode	= hns3_nic_set_rx_mode,
2226 	.ndo_vlan_rx_add_vid	= hns3_vlan_rx_add_vid,
2227 	.ndo_vlan_rx_kill_vid	= hns3_vlan_rx_kill_vid,
2228 	.ndo_set_vf_vlan	= hns3_ndo_set_vf_vlan,
2229 	.ndo_set_vf_spoofchk	= hns3_set_vf_spoofchk,
2230 	.ndo_set_vf_trust	= hns3_set_vf_trust,
2231 #ifdef CONFIG_RFS_ACCEL
2232 	.ndo_rx_flow_steer	= hns3_rx_flow_steer,
2233 #endif
2234 	.ndo_get_vf_config	= hns3_nic_get_vf_config,
2235 	.ndo_set_vf_link_state	= hns3_nic_set_vf_link_state,
2236 	.ndo_set_vf_rate	= hns3_nic_set_vf_rate,
2237 	.ndo_set_vf_mac		= hns3_nic_set_vf_mac,
2238 };
2239 
2240 bool hns3_is_phys_func(struct pci_dev *pdev)
2241 {
2242 	u32 dev_id = pdev->device;
2243 
2244 	switch (dev_id) {
2245 	case HNAE3_DEV_ID_GE:
2246 	case HNAE3_DEV_ID_25GE:
2247 	case HNAE3_DEV_ID_25GE_RDMA:
2248 	case HNAE3_DEV_ID_25GE_RDMA_MACSEC:
2249 	case HNAE3_DEV_ID_50GE_RDMA:
2250 	case HNAE3_DEV_ID_50GE_RDMA_MACSEC:
2251 	case HNAE3_DEV_ID_100G_RDMA_MACSEC:
2252 	case HNAE3_DEV_ID_200G_RDMA:
2253 		return true;
2254 	case HNAE3_DEV_ID_VF:
2255 	case HNAE3_DEV_ID_RDMA_DCB_PFC_VF:
2256 		return false;
2257 	default:
2258 		dev_warn(&pdev->dev, "un-recognized pci device-id %u",
2259 			 dev_id);
2260 	}
2261 
2262 	return false;
2263 }
2264 
2265 static void hns3_disable_sriov(struct pci_dev *pdev)
2266 {
2267 	/* If our VFs are assigned we cannot shut down SR-IOV
2268 	 * without causing issues, so just leave the hardware
2269 	 * available but disabled
2270 	 */
2271 	if (pci_vfs_assigned(pdev)) {
2272 		dev_warn(&pdev->dev,
2273 			 "disabling driver while VFs are assigned\n");
2274 		return;
2275 	}
2276 
2277 	pci_disable_sriov(pdev);
2278 }
2279 
2280 /* hns3_probe - Device initialization routine
2281  * @pdev: PCI device information struct
2282  * @ent: entry in hns3_pci_tbl
2283  *
2284  * hns3_probe initializes a PF identified by a pci_dev structure.
2285  * The OS initialization, configuring of the PF private structure,
2286  * and a hardware reset occur.
2287  *
2288  * Returns 0 on success, negative on failure
2289  */
2290 static int hns3_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2291 {
2292 	struct hnae3_ae_dev *ae_dev;
2293 	int ret;
2294 
2295 	ae_dev = devm_kzalloc(&pdev->dev, sizeof(*ae_dev), GFP_KERNEL);
2296 	if (!ae_dev)
2297 		return -ENOMEM;
2298 
2299 	ae_dev->pdev = pdev;
2300 	ae_dev->flag = ent->driver_data;
2301 	pci_set_drvdata(pdev, ae_dev);
2302 
2303 	ret = hnae3_register_ae_dev(ae_dev);
2304 	if (ret)
2305 		pci_set_drvdata(pdev, NULL);
2306 
2307 	return ret;
2308 }
2309 
2310 /* hns3_remove - Device removal routine
2311  * @pdev: PCI device information struct
2312  */
2313 static void hns3_remove(struct pci_dev *pdev)
2314 {
2315 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2316 
2317 	if (hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))
2318 		hns3_disable_sriov(pdev);
2319 
2320 	hnae3_unregister_ae_dev(ae_dev);
2321 	pci_set_drvdata(pdev, NULL);
2322 }
2323 
2324 /**
2325  * hns3_pci_sriov_configure
2326  * @pdev: pointer to a pci_dev structure
2327  * @num_vfs: number of VFs to allocate
2328  *
2329  * Enable or change the number of VFs. Called when the user updates the number
2330  * of VFs in sysfs.
2331  **/
2332 static int hns3_pci_sriov_configure(struct pci_dev *pdev, int num_vfs)
2333 {
2334 	int ret;
2335 
2336 	if (!(hns3_is_phys_func(pdev) && IS_ENABLED(CONFIG_PCI_IOV))) {
2337 		dev_warn(&pdev->dev, "Can not config SRIOV\n");
2338 		return -EINVAL;
2339 	}
2340 
2341 	if (num_vfs) {
2342 		ret = pci_enable_sriov(pdev, num_vfs);
2343 		if (ret)
2344 			dev_err(&pdev->dev, "SRIOV enable failed %d\n", ret);
2345 		else
2346 			return num_vfs;
2347 	} else if (!pci_vfs_assigned(pdev)) {
2348 		pci_disable_sriov(pdev);
2349 	} else {
2350 		dev_warn(&pdev->dev,
2351 			 "Unable to free VFs because some are assigned to VMs.\n");
2352 	}
2353 
2354 	return 0;
2355 }
2356 
2357 static void hns3_shutdown(struct pci_dev *pdev)
2358 {
2359 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2360 
2361 	hnae3_unregister_ae_dev(ae_dev);
2362 	pci_set_drvdata(pdev, NULL);
2363 
2364 	if (system_state == SYSTEM_POWER_OFF)
2365 		pci_set_power_state(pdev, PCI_D3hot);
2366 }
2367 
2368 static int __maybe_unused hns3_suspend(struct device *dev)
2369 {
2370 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
2371 
2372 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
2373 		dev_info(dev, "Begin to suspend.\n");
2374 		if (ae_dev->ops && ae_dev->ops->reset_prepare)
2375 			ae_dev->ops->reset_prepare(ae_dev, HNAE3_FUNC_RESET);
2376 	}
2377 
2378 	return 0;
2379 }
2380 
2381 static int __maybe_unused hns3_resume(struct device *dev)
2382 {
2383 	struct hnae3_ae_dev *ae_dev = dev_get_drvdata(dev);
2384 
2385 	if (ae_dev && hns3_is_phys_func(ae_dev->pdev)) {
2386 		dev_info(dev, "Begin to resume.\n");
2387 		if (ae_dev->ops && ae_dev->ops->reset_done)
2388 			ae_dev->ops->reset_done(ae_dev);
2389 	}
2390 
2391 	return 0;
2392 }
2393 
2394 static pci_ers_result_t hns3_error_detected(struct pci_dev *pdev,
2395 					    pci_channel_state_t state)
2396 {
2397 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2398 	pci_ers_result_t ret;
2399 
2400 	dev_info(&pdev->dev, "PCI error detected, state(=%u)!!\n", state);
2401 
2402 	if (state == pci_channel_io_perm_failure)
2403 		return PCI_ERS_RESULT_DISCONNECT;
2404 
2405 	if (!ae_dev || !ae_dev->ops) {
2406 		dev_err(&pdev->dev,
2407 			"Can't recover - error happened before device initialized\n");
2408 		return PCI_ERS_RESULT_NONE;
2409 	}
2410 
2411 	if (ae_dev->ops->handle_hw_ras_error)
2412 		ret = ae_dev->ops->handle_hw_ras_error(ae_dev);
2413 	else
2414 		return PCI_ERS_RESULT_NONE;
2415 
2416 	return ret;
2417 }
2418 
2419 static pci_ers_result_t hns3_slot_reset(struct pci_dev *pdev)
2420 {
2421 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2422 	const struct hnae3_ae_ops *ops;
2423 	enum hnae3_reset_type reset_type;
2424 	struct device *dev = &pdev->dev;
2425 
2426 	if (!ae_dev || !ae_dev->ops)
2427 		return PCI_ERS_RESULT_NONE;
2428 
2429 	ops = ae_dev->ops;
2430 	/* request the reset */
2431 	if (ops->reset_event && ops->get_reset_level &&
2432 	    ops->set_default_reset_request) {
2433 		if (ae_dev->hw_err_reset_req) {
2434 			reset_type = ops->get_reset_level(ae_dev,
2435 						&ae_dev->hw_err_reset_req);
2436 			ops->set_default_reset_request(ae_dev, reset_type);
2437 			dev_info(dev, "requesting reset due to PCI error\n");
2438 			ops->reset_event(pdev, NULL);
2439 		}
2440 
2441 		return PCI_ERS_RESULT_RECOVERED;
2442 	}
2443 
2444 	return PCI_ERS_RESULT_DISCONNECT;
2445 }
2446 
2447 static void hns3_reset_prepare(struct pci_dev *pdev)
2448 {
2449 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2450 
2451 	dev_info(&pdev->dev, "FLR prepare\n");
2452 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_prepare)
2453 		ae_dev->ops->reset_prepare(ae_dev, HNAE3_FLR_RESET);
2454 }
2455 
2456 static void hns3_reset_done(struct pci_dev *pdev)
2457 {
2458 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2459 
2460 	dev_info(&pdev->dev, "FLR done\n");
2461 	if (ae_dev && ae_dev->ops && ae_dev->ops->reset_done)
2462 		ae_dev->ops->reset_done(ae_dev);
2463 }
2464 
2465 static const struct pci_error_handlers hns3_err_handler = {
2466 	.error_detected = hns3_error_detected,
2467 	.slot_reset     = hns3_slot_reset,
2468 	.reset_prepare	= hns3_reset_prepare,
2469 	.reset_done	= hns3_reset_done,
2470 };
2471 
2472 static SIMPLE_DEV_PM_OPS(hns3_pm_ops, hns3_suspend, hns3_resume);
2473 
2474 static struct pci_driver hns3_driver = {
2475 	.name     = hns3_driver_name,
2476 	.id_table = hns3_pci_tbl,
2477 	.probe    = hns3_probe,
2478 	.remove   = hns3_remove,
2479 	.shutdown = hns3_shutdown,
2480 	.driver.pm  = &hns3_pm_ops,
2481 	.sriov_configure = hns3_pci_sriov_configure,
2482 	.err_handler    = &hns3_err_handler,
2483 };
2484 
2485 /* set default feature to hns3 */
2486 static void hns3_set_default_feature(struct net_device *netdev)
2487 {
2488 	struct hnae3_handle *h = hns3_get_handle(netdev);
2489 	struct pci_dev *pdev = h->pdev;
2490 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
2491 
2492 	netdev->priv_flags |= IFF_UNICAST_FLT;
2493 
2494 	netdev->hw_enc_features |= NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2495 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2496 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2497 		NETIF_F_SCTP_CRC | NETIF_F_TSO_MANGLEID | NETIF_F_FRAGLIST;
2498 
2499 	netdev->gso_partial_features |= NETIF_F_GSO_GRE_CSUM;
2500 
2501 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2502 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2503 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2504 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2505 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2506 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2507 
2508 	netdev->vlan_features |= NETIF_F_RXCSUM |
2509 		NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO |
2510 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2511 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2512 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2513 
2514 	netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX |
2515 		NETIF_F_HW_VLAN_CTAG_RX |
2516 		NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_GSO |
2517 		NETIF_F_GRO | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_GRE |
2518 		NETIF_F_GSO_GRE_CSUM | NETIF_F_GSO_UDP_TUNNEL |
2519 		NETIF_F_SCTP_CRC | NETIF_F_FRAGLIST;
2520 
2521 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V2) {
2522 		netdev->hw_features |= NETIF_F_GRO_HW;
2523 		netdev->features |= NETIF_F_GRO_HW;
2524 
2525 		if (!(h->flags & HNAE3_SUPPORT_VF)) {
2526 			netdev->hw_features |= NETIF_F_NTUPLE;
2527 			netdev->features |= NETIF_F_NTUPLE;
2528 		}
2529 	}
2530 
2531 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_GSO_B, ae_dev->caps)) {
2532 		netdev->hw_features |= NETIF_F_GSO_UDP_L4;
2533 		netdev->features |= NETIF_F_GSO_UDP_L4;
2534 		netdev->vlan_features |= NETIF_F_GSO_UDP_L4;
2535 		netdev->hw_enc_features |= NETIF_F_GSO_UDP_L4;
2536 	}
2537 
2538 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps)) {
2539 		netdev->hw_features |= NETIF_F_HW_CSUM;
2540 		netdev->features |= NETIF_F_HW_CSUM;
2541 		netdev->vlan_features |= NETIF_F_HW_CSUM;
2542 		netdev->hw_enc_features |= NETIF_F_HW_CSUM;
2543 	} else {
2544 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2545 		netdev->features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2546 		netdev->vlan_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2547 		netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2548 	}
2549 
2550 	if (test_bit(HNAE3_DEV_SUPPORT_UDP_TUNNEL_CSUM_B, ae_dev->caps)) {
2551 		netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2552 		netdev->features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2553 		netdev->vlan_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2554 		netdev->hw_enc_features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
2555 	}
2556 
2557 	if (test_bit(HNAE3_DEV_SUPPORT_FD_FORWARD_TC_B, ae_dev->caps)) {
2558 		netdev->hw_features |= NETIF_F_HW_TC;
2559 		netdev->features |= NETIF_F_HW_TC;
2560 	}
2561 }
2562 
2563 static int hns3_alloc_buffer(struct hns3_enet_ring *ring,
2564 			     struct hns3_desc_cb *cb)
2565 {
2566 	unsigned int order = hns3_page_order(ring);
2567 	struct page *p;
2568 
2569 	p = dev_alloc_pages(order);
2570 	if (!p)
2571 		return -ENOMEM;
2572 
2573 	cb->priv = p;
2574 	cb->page_offset = 0;
2575 	cb->reuse_flag = 0;
2576 	cb->buf  = page_address(p);
2577 	cb->length = hns3_page_size(ring);
2578 	cb->type = DESC_TYPE_PAGE;
2579 	page_ref_add(p, USHRT_MAX - 1);
2580 	cb->pagecnt_bias = USHRT_MAX;
2581 
2582 	return 0;
2583 }
2584 
2585 static void hns3_free_buffer(struct hns3_enet_ring *ring,
2586 			     struct hns3_desc_cb *cb, int budget)
2587 {
2588 	if (cb->type == DESC_TYPE_SKB)
2589 		napi_consume_skb(cb->priv, budget);
2590 	else if (!HNAE3_IS_TX_RING(ring) && cb->pagecnt_bias)
2591 		__page_frag_cache_drain(cb->priv, cb->pagecnt_bias);
2592 	memset(cb, 0, sizeof(*cb));
2593 }
2594 
2595 static int hns3_map_buffer(struct hns3_enet_ring *ring, struct hns3_desc_cb *cb)
2596 {
2597 	cb->dma = dma_map_page(ring_to_dev(ring), cb->priv, 0,
2598 			       cb->length, ring_to_dma_dir(ring));
2599 
2600 	if (unlikely(dma_mapping_error(ring_to_dev(ring), cb->dma)))
2601 		return -EIO;
2602 
2603 	return 0;
2604 }
2605 
2606 static void hns3_unmap_buffer(struct hns3_enet_ring *ring,
2607 			      struct hns3_desc_cb *cb)
2608 {
2609 	if (cb->type == DESC_TYPE_SKB || cb->type == DESC_TYPE_FRAGLIST_SKB)
2610 		dma_unmap_single(ring_to_dev(ring), cb->dma, cb->length,
2611 				 ring_to_dma_dir(ring));
2612 	else if (cb->length)
2613 		dma_unmap_page(ring_to_dev(ring), cb->dma, cb->length,
2614 			       ring_to_dma_dir(ring));
2615 }
2616 
2617 static void hns3_buffer_detach(struct hns3_enet_ring *ring, int i)
2618 {
2619 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2620 	ring->desc[i].addr = 0;
2621 }
2622 
2623 static void hns3_free_buffer_detach(struct hns3_enet_ring *ring, int i,
2624 				    int budget)
2625 {
2626 	struct hns3_desc_cb *cb = &ring->desc_cb[i];
2627 
2628 	if (!ring->desc_cb[i].dma)
2629 		return;
2630 
2631 	hns3_buffer_detach(ring, i);
2632 	hns3_free_buffer(ring, cb, budget);
2633 }
2634 
2635 static void hns3_free_buffers(struct hns3_enet_ring *ring)
2636 {
2637 	int i;
2638 
2639 	for (i = 0; i < ring->desc_num; i++)
2640 		hns3_free_buffer_detach(ring, i, 0);
2641 }
2642 
2643 /* free desc along with its attached buffer */
2644 static void hns3_free_desc(struct hns3_enet_ring *ring)
2645 {
2646 	int size = ring->desc_num * sizeof(ring->desc[0]);
2647 
2648 	hns3_free_buffers(ring);
2649 
2650 	if (ring->desc) {
2651 		dma_free_coherent(ring_to_dev(ring), size,
2652 				  ring->desc, ring->desc_dma_addr);
2653 		ring->desc = NULL;
2654 	}
2655 }
2656 
2657 static int hns3_alloc_desc(struct hns3_enet_ring *ring)
2658 {
2659 	int size = ring->desc_num * sizeof(ring->desc[0]);
2660 
2661 	ring->desc = dma_alloc_coherent(ring_to_dev(ring), size,
2662 					&ring->desc_dma_addr, GFP_KERNEL);
2663 	if (!ring->desc)
2664 		return -ENOMEM;
2665 
2666 	return 0;
2667 }
2668 
2669 static int hns3_alloc_and_map_buffer(struct hns3_enet_ring *ring,
2670 				   struct hns3_desc_cb *cb)
2671 {
2672 	int ret;
2673 
2674 	ret = hns3_alloc_buffer(ring, cb);
2675 	if (ret)
2676 		goto out;
2677 
2678 	ret = hns3_map_buffer(ring, cb);
2679 	if (ret)
2680 		goto out_with_buf;
2681 
2682 	return 0;
2683 
2684 out_with_buf:
2685 	hns3_free_buffer(ring, cb, 0);
2686 out:
2687 	return ret;
2688 }
2689 
2690 static int hns3_alloc_and_attach_buffer(struct hns3_enet_ring *ring, int i)
2691 {
2692 	int ret = hns3_alloc_and_map_buffer(ring, &ring->desc_cb[i]);
2693 
2694 	if (ret)
2695 		return ret;
2696 
2697 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2698 
2699 	return 0;
2700 }
2701 
2702 /* Allocate memory for raw pkg, and map with dma */
2703 static int hns3_alloc_ring_buffers(struct hns3_enet_ring *ring)
2704 {
2705 	int i, j, ret;
2706 
2707 	for (i = 0; i < ring->desc_num; i++) {
2708 		ret = hns3_alloc_and_attach_buffer(ring, i);
2709 		if (ret)
2710 			goto out_buffer_fail;
2711 	}
2712 
2713 	return 0;
2714 
2715 out_buffer_fail:
2716 	for (j = i - 1; j >= 0; j--)
2717 		hns3_free_buffer_detach(ring, j, 0);
2718 	return ret;
2719 }
2720 
2721 /* detach a in-used buffer and replace with a reserved one */
2722 static void hns3_replace_buffer(struct hns3_enet_ring *ring, int i,
2723 				struct hns3_desc_cb *res_cb)
2724 {
2725 	hns3_unmap_buffer(ring, &ring->desc_cb[i]);
2726 	ring->desc_cb[i] = *res_cb;
2727 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma);
2728 	ring->desc[i].rx.bd_base_info = 0;
2729 }
2730 
2731 static void hns3_reuse_buffer(struct hns3_enet_ring *ring, int i)
2732 {
2733 	ring->desc_cb[i].reuse_flag = 0;
2734 	ring->desc[i].addr = cpu_to_le64(ring->desc_cb[i].dma +
2735 					 ring->desc_cb[i].page_offset);
2736 	ring->desc[i].rx.bd_base_info = 0;
2737 
2738 	dma_sync_single_for_device(ring_to_dev(ring),
2739 			ring->desc_cb[i].dma + ring->desc_cb[i].page_offset,
2740 			hns3_buf_size(ring),
2741 			DMA_FROM_DEVICE);
2742 }
2743 
2744 static bool hns3_nic_reclaim_desc(struct hns3_enet_ring *ring,
2745 				  int *bytes, int *pkts, int budget)
2746 {
2747 	/* pair with ring->last_to_use update in hns3_tx_doorbell(),
2748 	 * smp_store_release() is not used in hns3_tx_doorbell() because
2749 	 * the doorbell operation already have the needed barrier operation.
2750 	 */
2751 	int ltu = smp_load_acquire(&ring->last_to_use);
2752 	int ntc = ring->next_to_clean;
2753 	struct hns3_desc_cb *desc_cb;
2754 	bool reclaimed = false;
2755 	struct hns3_desc *desc;
2756 
2757 	while (ltu != ntc) {
2758 		desc = &ring->desc[ntc];
2759 
2760 		if (le16_to_cpu(desc->tx.bdtp_fe_sc_vld_ra_ri) &
2761 				BIT(HNS3_TXD_VLD_B))
2762 			break;
2763 
2764 		desc_cb = &ring->desc_cb[ntc];
2765 
2766 		if (desc_cb->type == DESC_TYPE_SKB) {
2767 			(*pkts)++;
2768 			(*bytes) += desc_cb->send_bytes;
2769 		}
2770 
2771 		/* desc_cb will be cleaned, after hnae3_free_buffer_detach */
2772 		hns3_free_buffer_detach(ring, ntc, budget);
2773 
2774 		if (++ntc == ring->desc_num)
2775 			ntc = 0;
2776 
2777 		/* Issue prefetch for next Tx descriptor */
2778 		prefetch(&ring->desc_cb[ntc]);
2779 		reclaimed = true;
2780 	}
2781 
2782 	if (unlikely(!reclaimed))
2783 		return false;
2784 
2785 	/* This smp_store_release() pairs with smp_load_acquire() in
2786 	 * ring_space called by hns3_nic_net_xmit.
2787 	 */
2788 	smp_store_release(&ring->next_to_clean, ntc);
2789 	return true;
2790 }
2791 
2792 void hns3_clean_tx_ring(struct hns3_enet_ring *ring, int budget)
2793 {
2794 	struct net_device *netdev = ring_to_netdev(ring);
2795 	struct hns3_nic_priv *priv = netdev_priv(netdev);
2796 	struct netdev_queue *dev_queue;
2797 	int bytes, pkts;
2798 
2799 	bytes = 0;
2800 	pkts = 0;
2801 
2802 	if (unlikely(!hns3_nic_reclaim_desc(ring, &bytes, &pkts, budget)))
2803 		return;
2804 
2805 	ring->tqp_vector->tx_group.total_bytes += bytes;
2806 	ring->tqp_vector->tx_group.total_packets += pkts;
2807 
2808 	u64_stats_update_begin(&ring->syncp);
2809 	ring->stats.tx_bytes += bytes;
2810 	ring->stats.tx_pkts += pkts;
2811 	u64_stats_update_end(&ring->syncp);
2812 
2813 	dev_queue = netdev_get_tx_queue(netdev, ring->tqp->tqp_index);
2814 	netdev_tx_completed_queue(dev_queue, pkts, bytes);
2815 
2816 	if (unlikely(netif_carrier_ok(netdev) &&
2817 		     ring_space(ring) > HNS3_MAX_TSO_BD_NUM)) {
2818 		/* Make sure that anybody stopping the queue after this
2819 		 * sees the new next_to_clean.
2820 		 */
2821 		smp_mb();
2822 		if (netif_tx_queue_stopped(dev_queue) &&
2823 		    !test_bit(HNS3_NIC_STATE_DOWN, &priv->state)) {
2824 			netif_tx_wake_queue(dev_queue);
2825 			ring->stats.restart_queue++;
2826 		}
2827 	}
2828 }
2829 
2830 static int hns3_desc_unused(struct hns3_enet_ring *ring)
2831 {
2832 	int ntc = ring->next_to_clean;
2833 	int ntu = ring->next_to_use;
2834 
2835 	return ((ntc >= ntu) ? 0 : ring->desc_num) + ntc - ntu;
2836 }
2837 
2838 static void hns3_nic_alloc_rx_buffers(struct hns3_enet_ring *ring,
2839 				      int cleand_count)
2840 {
2841 	struct hns3_desc_cb *desc_cb;
2842 	struct hns3_desc_cb res_cbs;
2843 	int i, ret;
2844 
2845 	for (i = 0; i < cleand_count; i++) {
2846 		desc_cb = &ring->desc_cb[ring->next_to_use];
2847 		if (desc_cb->reuse_flag) {
2848 			u64_stats_update_begin(&ring->syncp);
2849 			ring->stats.reuse_pg_cnt++;
2850 			u64_stats_update_end(&ring->syncp);
2851 
2852 			hns3_reuse_buffer(ring, ring->next_to_use);
2853 		} else {
2854 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
2855 			if (ret) {
2856 				u64_stats_update_begin(&ring->syncp);
2857 				ring->stats.sw_err_cnt++;
2858 				u64_stats_update_end(&ring->syncp);
2859 
2860 				hns3_rl_err(ring_to_netdev(ring),
2861 					    "alloc rx buffer failed: %d\n",
2862 					    ret);
2863 				break;
2864 			}
2865 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
2866 
2867 			u64_stats_update_begin(&ring->syncp);
2868 			ring->stats.non_reuse_pg++;
2869 			u64_stats_update_end(&ring->syncp);
2870 		}
2871 
2872 		ring_ptr_move_fw(ring, next_to_use);
2873 	}
2874 
2875 	writel(i, ring->tqp->io_base + HNS3_RING_RX_RING_HEAD_REG);
2876 }
2877 
2878 static bool hns3_can_reuse_page(struct hns3_desc_cb *cb)
2879 {
2880 	return (page_count(cb->priv) - cb->pagecnt_bias) == 1;
2881 }
2882 
2883 static void hns3_nic_reuse_page(struct sk_buff *skb, int i,
2884 				struct hns3_enet_ring *ring, int pull_len,
2885 				struct hns3_desc_cb *desc_cb)
2886 {
2887 	struct hns3_desc *desc = &ring->desc[ring->next_to_clean];
2888 	int size = le16_to_cpu(desc->rx.size);
2889 	u32 truesize = hns3_buf_size(ring);
2890 
2891 	desc_cb->pagecnt_bias--;
2892 	skb_add_rx_frag(skb, i, desc_cb->priv, desc_cb->page_offset + pull_len,
2893 			size - pull_len, truesize);
2894 
2895 	/* Avoid re-using remote and pfmemalloc pages, or the stack is still
2896 	 * using the page when page_offset rollback to zero, flag default
2897 	 * unreuse
2898 	 */
2899 	if (!dev_page_is_reusable(desc_cb->priv) ||
2900 	    (!desc_cb->page_offset && !hns3_can_reuse_page(desc_cb))) {
2901 		__page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2902 		return;
2903 	}
2904 
2905 	/* Move offset up to the next cache line */
2906 	desc_cb->page_offset += truesize;
2907 
2908 	if (desc_cb->page_offset + truesize <= hns3_page_size(ring)) {
2909 		desc_cb->reuse_flag = 1;
2910 	} else if (hns3_can_reuse_page(desc_cb)) {
2911 		desc_cb->reuse_flag = 1;
2912 		desc_cb->page_offset = 0;
2913 	} else if (desc_cb->pagecnt_bias) {
2914 		__page_frag_cache_drain(desc_cb->priv, desc_cb->pagecnt_bias);
2915 		return;
2916 	}
2917 
2918 	if (unlikely(!desc_cb->pagecnt_bias)) {
2919 		page_ref_add(desc_cb->priv, USHRT_MAX);
2920 		desc_cb->pagecnt_bias = USHRT_MAX;
2921 	}
2922 }
2923 
2924 static int hns3_gro_complete(struct sk_buff *skb, u32 l234info)
2925 {
2926 	__be16 type = skb->protocol;
2927 	struct tcphdr *th;
2928 	int depth = 0;
2929 
2930 	while (eth_type_vlan(type)) {
2931 		struct vlan_hdr *vh;
2932 
2933 		if ((depth + VLAN_HLEN) > skb_headlen(skb))
2934 			return -EFAULT;
2935 
2936 		vh = (struct vlan_hdr *)(skb->data + depth);
2937 		type = vh->h_vlan_encapsulated_proto;
2938 		depth += VLAN_HLEN;
2939 	}
2940 
2941 	skb_set_network_header(skb, depth);
2942 
2943 	if (type == htons(ETH_P_IP)) {
2944 		const struct iphdr *iph = ip_hdr(skb);
2945 
2946 		depth += sizeof(struct iphdr);
2947 		skb_set_transport_header(skb, depth);
2948 		th = tcp_hdr(skb);
2949 		th->check = ~tcp_v4_check(skb->len - depth, iph->saddr,
2950 					  iph->daddr, 0);
2951 	} else if (type == htons(ETH_P_IPV6)) {
2952 		const struct ipv6hdr *iph = ipv6_hdr(skb);
2953 
2954 		depth += sizeof(struct ipv6hdr);
2955 		skb_set_transport_header(skb, depth);
2956 		th = tcp_hdr(skb);
2957 		th->check = ~tcp_v6_check(skb->len - depth, &iph->saddr,
2958 					  &iph->daddr, 0);
2959 	} else {
2960 		hns3_rl_err(skb->dev,
2961 			    "Error: FW GRO supports only IPv4/IPv6, not 0x%04x, depth: %d\n",
2962 			    be16_to_cpu(type), depth);
2963 		return -EFAULT;
2964 	}
2965 
2966 	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
2967 	if (th->cwr)
2968 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
2969 
2970 	if (l234info & BIT(HNS3_RXD_GRO_FIXID_B))
2971 		skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_FIXEDID;
2972 
2973 	skb->csum_start = (unsigned char *)th - skb->head;
2974 	skb->csum_offset = offsetof(struct tcphdr, check);
2975 	skb->ip_summed = CHECKSUM_PARTIAL;
2976 
2977 	trace_hns3_gro(skb);
2978 
2979 	return 0;
2980 }
2981 
2982 static void hns3_checksum_complete(struct hns3_enet_ring *ring,
2983 				   struct sk_buff *skb, u32 l234info)
2984 {
2985 	u32 lo, hi;
2986 
2987 	u64_stats_update_begin(&ring->syncp);
2988 	ring->stats.csum_complete++;
2989 	u64_stats_update_end(&ring->syncp);
2990 	skb->ip_summed = CHECKSUM_COMPLETE;
2991 	lo = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_L_M,
2992 			     HNS3_RXD_L2_CSUM_L_S);
2993 	hi = hnae3_get_field(l234info, HNS3_RXD_L2_CSUM_H_M,
2994 			     HNS3_RXD_L2_CSUM_H_S);
2995 	skb->csum = csum_unfold((__force __sum16)(lo | hi << 8));
2996 }
2997 
2998 static void hns3_rx_checksum(struct hns3_enet_ring *ring, struct sk_buff *skb,
2999 			     u32 l234info, u32 bd_base_info, u32 ol_info)
3000 {
3001 	struct net_device *netdev = ring_to_netdev(ring);
3002 	int l3_type, l4_type;
3003 	int ol4_type;
3004 
3005 	skb->ip_summed = CHECKSUM_NONE;
3006 
3007 	skb_checksum_none_assert(skb);
3008 
3009 	if (!(netdev->features & NETIF_F_RXCSUM))
3010 		return;
3011 
3012 	if (l234info & BIT(HNS3_RXD_L2_CSUM_B)) {
3013 		hns3_checksum_complete(ring, skb, l234info);
3014 		return;
3015 	}
3016 
3017 	/* check if hardware has done checksum */
3018 	if (!(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
3019 		return;
3020 
3021 	if (unlikely(l234info & (BIT(HNS3_RXD_L3E_B) | BIT(HNS3_RXD_L4E_B) |
3022 				 BIT(HNS3_RXD_OL3E_B) |
3023 				 BIT(HNS3_RXD_OL4E_B)))) {
3024 		u64_stats_update_begin(&ring->syncp);
3025 		ring->stats.l3l4_csum_err++;
3026 		u64_stats_update_end(&ring->syncp);
3027 
3028 		return;
3029 	}
3030 
3031 	ol4_type = hnae3_get_field(ol_info, HNS3_RXD_OL4ID_M,
3032 				   HNS3_RXD_OL4ID_S);
3033 	switch (ol4_type) {
3034 	case HNS3_OL4_TYPE_MAC_IN_UDP:
3035 	case HNS3_OL4_TYPE_NVGRE:
3036 		skb->csum_level = 1;
3037 		fallthrough;
3038 	case HNS3_OL4_TYPE_NO_TUN:
3039 		l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M,
3040 					  HNS3_RXD_L3ID_S);
3041 		l4_type = hnae3_get_field(l234info, HNS3_RXD_L4ID_M,
3042 					  HNS3_RXD_L4ID_S);
3043 		/* Can checksum ipv4 or ipv6 + UDP/TCP/SCTP packets */
3044 		if ((l3_type == HNS3_L3_TYPE_IPV4 ||
3045 		     l3_type == HNS3_L3_TYPE_IPV6) &&
3046 		    (l4_type == HNS3_L4_TYPE_UDP ||
3047 		     l4_type == HNS3_L4_TYPE_TCP ||
3048 		     l4_type == HNS3_L4_TYPE_SCTP))
3049 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3050 		break;
3051 	default:
3052 		break;
3053 	}
3054 }
3055 
3056 static void hns3_rx_skb(struct hns3_enet_ring *ring, struct sk_buff *skb)
3057 {
3058 	if (skb_has_frag_list(skb))
3059 		napi_gro_flush(&ring->tqp_vector->napi, false);
3060 
3061 	napi_gro_receive(&ring->tqp_vector->napi, skb);
3062 }
3063 
3064 static bool hns3_parse_vlan_tag(struct hns3_enet_ring *ring,
3065 				struct hns3_desc *desc, u32 l234info,
3066 				u16 *vlan_tag)
3067 {
3068 	struct hnae3_handle *handle = ring->tqp->handle;
3069 	struct pci_dev *pdev = ring->tqp->handle->pdev;
3070 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
3071 
3072 	if (unlikely(ae_dev->dev_version < HNAE3_DEVICE_VERSION_V2)) {
3073 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3074 		if (!(*vlan_tag & VLAN_VID_MASK))
3075 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3076 
3077 		return (*vlan_tag != 0);
3078 	}
3079 
3080 #define HNS3_STRP_OUTER_VLAN	0x1
3081 #define HNS3_STRP_INNER_VLAN	0x2
3082 #define HNS3_STRP_BOTH		0x3
3083 
3084 	/* Hardware always insert VLAN tag into RX descriptor when
3085 	 * remove the tag from packet, driver needs to determine
3086 	 * reporting which tag to stack.
3087 	 */
3088 	switch (hnae3_get_field(l234info, HNS3_RXD_STRP_TAGP_M,
3089 				HNS3_RXD_STRP_TAGP_S)) {
3090 	case HNS3_STRP_OUTER_VLAN:
3091 		if (handle->port_base_vlan_state !=
3092 				HNAE3_PORT_BASE_VLAN_DISABLE)
3093 			return false;
3094 
3095 		*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3096 		return true;
3097 	case HNS3_STRP_INNER_VLAN:
3098 		if (handle->port_base_vlan_state !=
3099 				HNAE3_PORT_BASE_VLAN_DISABLE)
3100 			return false;
3101 
3102 		*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3103 		return true;
3104 	case HNS3_STRP_BOTH:
3105 		if (handle->port_base_vlan_state ==
3106 				HNAE3_PORT_BASE_VLAN_DISABLE)
3107 			*vlan_tag = le16_to_cpu(desc->rx.ot_vlan_tag);
3108 		else
3109 			*vlan_tag = le16_to_cpu(desc->rx.vlan_tag);
3110 
3111 		return true;
3112 	default:
3113 		return false;
3114 	}
3115 }
3116 
3117 static void hns3_rx_ring_move_fw(struct hns3_enet_ring *ring)
3118 {
3119 	ring->desc[ring->next_to_clean].rx.bd_base_info &=
3120 		cpu_to_le32(~BIT(HNS3_RXD_VLD_B));
3121 	ring->next_to_clean += 1;
3122 
3123 	if (unlikely(ring->next_to_clean == ring->desc_num))
3124 		ring->next_to_clean = 0;
3125 }
3126 
3127 static int hns3_alloc_skb(struct hns3_enet_ring *ring, unsigned int length,
3128 			  unsigned char *va)
3129 {
3130 	struct hns3_desc_cb *desc_cb = &ring->desc_cb[ring->next_to_clean];
3131 	struct net_device *netdev = ring_to_netdev(ring);
3132 	struct sk_buff *skb;
3133 
3134 	ring->skb = napi_alloc_skb(&ring->tqp_vector->napi, HNS3_RX_HEAD_SIZE);
3135 	skb = ring->skb;
3136 	if (unlikely(!skb)) {
3137 		hns3_rl_err(netdev, "alloc rx skb fail\n");
3138 
3139 		u64_stats_update_begin(&ring->syncp);
3140 		ring->stats.sw_err_cnt++;
3141 		u64_stats_update_end(&ring->syncp);
3142 
3143 		return -ENOMEM;
3144 	}
3145 
3146 	trace_hns3_rx_desc(ring);
3147 	prefetchw(skb->data);
3148 
3149 	ring->pending_buf = 1;
3150 	ring->frag_num = 0;
3151 	ring->tail_skb = NULL;
3152 	if (length <= HNS3_RX_HEAD_SIZE) {
3153 		memcpy(__skb_put(skb, length), va, ALIGN(length, sizeof(long)));
3154 
3155 		/* We can reuse buffer as-is, just make sure it is reusable */
3156 		if (dev_page_is_reusable(desc_cb->priv))
3157 			desc_cb->reuse_flag = 1;
3158 		else /* This page cannot be reused so discard it */
3159 			__page_frag_cache_drain(desc_cb->priv,
3160 						desc_cb->pagecnt_bias);
3161 
3162 		hns3_rx_ring_move_fw(ring);
3163 		return 0;
3164 	}
3165 	u64_stats_update_begin(&ring->syncp);
3166 	ring->stats.seg_pkt_cnt++;
3167 	u64_stats_update_end(&ring->syncp);
3168 
3169 	ring->pull_len = eth_get_headlen(netdev, va, HNS3_RX_HEAD_SIZE);
3170 	__skb_put(skb, ring->pull_len);
3171 	hns3_nic_reuse_page(skb, ring->frag_num++, ring, ring->pull_len,
3172 			    desc_cb);
3173 	hns3_rx_ring_move_fw(ring);
3174 
3175 	return 0;
3176 }
3177 
3178 static int hns3_add_frag(struct hns3_enet_ring *ring)
3179 {
3180 	struct sk_buff *skb = ring->skb;
3181 	struct sk_buff *head_skb = skb;
3182 	struct sk_buff *new_skb;
3183 	struct hns3_desc_cb *desc_cb;
3184 	struct hns3_desc *desc;
3185 	u32 bd_base_info;
3186 
3187 	do {
3188 		desc = &ring->desc[ring->next_to_clean];
3189 		desc_cb = &ring->desc_cb[ring->next_to_clean];
3190 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3191 		/* make sure HW write desc complete */
3192 		dma_rmb();
3193 		if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
3194 			return -ENXIO;
3195 
3196 		if (unlikely(ring->frag_num >= MAX_SKB_FRAGS)) {
3197 			new_skb = napi_alloc_skb(&ring->tqp_vector->napi, 0);
3198 			if (unlikely(!new_skb)) {
3199 				hns3_rl_err(ring_to_netdev(ring),
3200 					    "alloc rx fraglist skb fail\n");
3201 				return -ENXIO;
3202 			}
3203 			ring->frag_num = 0;
3204 
3205 			if (ring->tail_skb) {
3206 				ring->tail_skb->next = new_skb;
3207 				ring->tail_skb = new_skb;
3208 			} else {
3209 				skb_shinfo(skb)->frag_list = new_skb;
3210 				ring->tail_skb = new_skb;
3211 			}
3212 		}
3213 
3214 		if (ring->tail_skb) {
3215 			head_skb->truesize += hns3_buf_size(ring);
3216 			head_skb->data_len += le16_to_cpu(desc->rx.size);
3217 			head_skb->len += le16_to_cpu(desc->rx.size);
3218 			skb = ring->tail_skb;
3219 		}
3220 
3221 		dma_sync_single_for_cpu(ring_to_dev(ring),
3222 				desc_cb->dma + desc_cb->page_offset,
3223 				hns3_buf_size(ring),
3224 				DMA_FROM_DEVICE);
3225 
3226 		hns3_nic_reuse_page(skb, ring->frag_num++, ring, 0, desc_cb);
3227 		trace_hns3_rx_desc(ring);
3228 		hns3_rx_ring_move_fw(ring);
3229 		ring->pending_buf++;
3230 	} while (!(bd_base_info & BIT(HNS3_RXD_FE_B)));
3231 
3232 	return 0;
3233 }
3234 
3235 static int hns3_set_gro_and_checksum(struct hns3_enet_ring *ring,
3236 				     struct sk_buff *skb, u32 l234info,
3237 				     u32 bd_base_info, u32 ol_info)
3238 {
3239 	u32 l3_type;
3240 
3241 	skb_shinfo(skb)->gso_size = hnae3_get_field(bd_base_info,
3242 						    HNS3_RXD_GRO_SIZE_M,
3243 						    HNS3_RXD_GRO_SIZE_S);
3244 	/* if there is no HW GRO, do not set gro params */
3245 	if (!skb_shinfo(skb)->gso_size) {
3246 		hns3_rx_checksum(ring, skb, l234info, bd_base_info, ol_info);
3247 		return 0;
3248 	}
3249 
3250 	NAPI_GRO_CB(skb)->count = hnae3_get_field(l234info,
3251 						  HNS3_RXD_GRO_COUNT_M,
3252 						  HNS3_RXD_GRO_COUNT_S);
3253 
3254 	l3_type = hnae3_get_field(l234info, HNS3_RXD_L3ID_M, HNS3_RXD_L3ID_S);
3255 	if (l3_type == HNS3_L3_TYPE_IPV4)
3256 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
3257 	else if (l3_type == HNS3_L3_TYPE_IPV6)
3258 		skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
3259 	else
3260 		return -EFAULT;
3261 
3262 	return  hns3_gro_complete(skb, l234info);
3263 }
3264 
3265 static void hns3_set_rx_skb_rss_type(struct hns3_enet_ring *ring,
3266 				     struct sk_buff *skb, u32 rss_hash)
3267 {
3268 	struct hnae3_handle *handle = ring->tqp->handle;
3269 	enum pkt_hash_types rss_type;
3270 
3271 	if (rss_hash)
3272 		rss_type = handle->kinfo.rss_type;
3273 	else
3274 		rss_type = PKT_HASH_TYPE_NONE;
3275 
3276 	skb_set_hash(skb, rss_hash, rss_type);
3277 }
3278 
3279 static int hns3_handle_bdinfo(struct hns3_enet_ring *ring, struct sk_buff *skb)
3280 {
3281 	struct net_device *netdev = ring_to_netdev(ring);
3282 	enum hns3_pkt_l2t_type l2_frame_type;
3283 	u32 bd_base_info, l234info, ol_info;
3284 	struct hns3_desc *desc;
3285 	unsigned int len;
3286 	int pre_ntc, ret;
3287 
3288 	/* bdinfo handled below is only valid on the last BD of the
3289 	 * current packet, and ring->next_to_clean indicates the first
3290 	 * descriptor of next packet, so need - 1 below.
3291 	 */
3292 	pre_ntc = ring->next_to_clean ? (ring->next_to_clean - 1) :
3293 					(ring->desc_num - 1);
3294 	desc = &ring->desc[pre_ntc];
3295 	bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3296 	l234info = le32_to_cpu(desc->rx.l234_info);
3297 	ol_info = le32_to_cpu(desc->rx.ol_info);
3298 
3299 	/* Based on hw strategy, the tag offloaded will be stored at
3300 	 * ot_vlan_tag in two layer tag case, and stored at vlan_tag
3301 	 * in one layer tag case.
3302 	 */
3303 	if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
3304 		u16 vlan_tag;
3305 
3306 		if (hns3_parse_vlan_tag(ring, desc, l234info, &vlan_tag))
3307 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3308 					       vlan_tag);
3309 	}
3310 
3311 	if (unlikely(!desc->rx.pkt_len || (l234info & (BIT(HNS3_RXD_TRUNCAT_B) |
3312 				  BIT(HNS3_RXD_L2E_B))))) {
3313 		u64_stats_update_begin(&ring->syncp);
3314 		if (l234info & BIT(HNS3_RXD_L2E_B))
3315 			ring->stats.l2_err++;
3316 		else
3317 			ring->stats.err_pkt_len++;
3318 		u64_stats_update_end(&ring->syncp);
3319 
3320 		return -EFAULT;
3321 	}
3322 
3323 	len = skb->len;
3324 
3325 	/* Do update ip stack process */
3326 	skb->protocol = eth_type_trans(skb, netdev);
3327 
3328 	/* This is needed in order to enable forwarding support */
3329 	ret = hns3_set_gro_and_checksum(ring, skb, l234info,
3330 					bd_base_info, ol_info);
3331 	if (unlikely(ret)) {
3332 		u64_stats_update_begin(&ring->syncp);
3333 		ring->stats.rx_err_cnt++;
3334 		u64_stats_update_end(&ring->syncp);
3335 		return ret;
3336 	}
3337 
3338 	l2_frame_type = hnae3_get_field(l234info, HNS3_RXD_DMAC_M,
3339 					HNS3_RXD_DMAC_S);
3340 
3341 	u64_stats_update_begin(&ring->syncp);
3342 	ring->stats.rx_pkts++;
3343 	ring->stats.rx_bytes += len;
3344 
3345 	if (l2_frame_type == HNS3_L2_TYPE_MULTICAST)
3346 		ring->stats.rx_multicast++;
3347 
3348 	u64_stats_update_end(&ring->syncp);
3349 
3350 	ring->tqp_vector->rx_group.total_bytes += len;
3351 
3352 	hns3_set_rx_skb_rss_type(ring, skb, le32_to_cpu(desc->rx.rss_hash));
3353 	return 0;
3354 }
3355 
3356 static int hns3_handle_rx_bd(struct hns3_enet_ring *ring)
3357 {
3358 	struct sk_buff *skb = ring->skb;
3359 	struct hns3_desc_cb *desc_cb;
3360 	struct hns3_desc *desc;
3361 	unsigned int length;
3362 	u32 bd_base_info;
3363 	int ret;
3364 
3365 	desc = &ring->desc[ring->next_to_clean];
3366 	desc_cb = &ring->desc_cb[ring->next_to_clean];
3367 
3368 	prefetch(desc);
3369 
3370 	if (!skb) {
3371 		bd_base_info = le32_to_cpu(desc->rx.bd_base_info);
3372 		/* Check valid BD */
3373 		if (unlikely(!(bd_base_info & BIT(HNS3_RXD_VLD_B))))
3374 			return -ENXIO;
3375 
3376 		dma_rmb();
3377 		length = le16_to_cpu(desc->rx.size);
3378 
3379 		ring->va = desc_cb->buf + desc_cb->page_offset;
3380 
3381 		dma_sync_single_for_cpu(ring_to_dev(ring),
3382 				desc_cb->dma + desc_cb->page_offset,
3383 				hns3_buf_size(ring),
3384 				DMA_FROM_DEVICE);
3385 
3386 		/* Prefetch first cache line of first page.
3387 		 * Idea is to cache few bytes of the header of the packet.
3388 		 * Our L1 Cache line size is 64B so need to prefetch twice to make
3389 		 * it 128B. But in actual we can have greater size of caches with
3390 		 * 128B Level 1 cache lines. In such a case, single fetch would
3391 		 * suffice to cache in the relevant part of the header.
3392 		 */
3393 		net_prefetch(ring->va);
3394 
3395 		ret = hns3_alloc_skb(ring, length, ring->va);
3396 		skb = ring->skb;
3397 
3398 		if (ret < 0) /* alloc buffer fail */
3399 			return ret;
3400 		if (!(bd_base_info & BIT(HNS3_RXD_FE_B))) { /* need add frag */
3401 			ret = hns3_add_frag(ring);
3402 			if (ret)
3403 				return ret;
3404 		}
3405 	} else {
3406 		ret = hns3_add_frag(ring);
3407 		if (ret)
3408 			return ret;
3409 	}
3410 
3411 	/* As the head data may be changed when GRO enable, copy
3412 	 * the head data in after other data rx completed
3413 	 */
3414 	if (skb->len > HNS3_RX_HEAD_SIZE)
3415 		memcpy(skb->data, ring->va,
3416 		       ALIGN(ring->pull_len, sizeof(long)));
3417 
3418 	ret = hns3_handle_bdinfo(ring, skb);
3419 	if (unlikely(ret)) {
3420 		dev_kfree_skb_any(skb);
3421 		return ret;
3422 	}
3423 
3424 	skb_record_rx_queue(skb, ring->tqp->tqp_index);
3425 	return 0;
3426 }
3427 
3428 int hns3_clean_rx_ring(struct hns3_enet_ring *ring, int budget,
3429 		       void (*rx_fn)(struct hns3_enet_ring *, struct sk_buff *))
3430 {
3431 #define RCB_NOF_ALLOC_RX_BUFF_ONCE 16
3432 	int unused_count = hns3_desc_unused(ring);
3433 	int recv_pkts = 0;
3434 	int err;
3435 
3436 	unused_count -= ring->pending_buf;
3437 
3438 	while (recv_pkts < budget) {
3439 		/* Reuse or realloc buffers */
3440 		if (unused_count >= RCB_NOF_ALLOC_RX_BUFF_ONCE) {
3441 			hns3_nic_alloc_rx_buffers(ring, unused_count);
3442 			unused_count = hns3_desc_unused(ring) -
3443 					ring->pending_buf;
3444 		}
3445 
3446 		/* Poll one pkt */
3447 		err = hns3_handle_rx_bd(ring);
3448 		/* Do not get FE for the packet or failed to alloc skb */
3449 		if (unlikely(!ring->skb || err == -ENXIO)) {
3450 			goto out;
3451 		} else if (likely(!err)) {
3452 			rx_fn(ring, ring->skb);
3453 			recv_pkts++;
3454 		}
3455 
3456 		unused_count += ring->pending_buf;
3457 		ring->skb = NULL;
3458 		ring->pending_buf = 0;
3459 	}
3460 
3461 out:
3462 	/* Make all data has been write before submit */
3463 	if (unused_count > 0)
3464 		hns3_nic_alloc_rx_buffers(ring, unused_count);
3465 
3466 	return recv_pkts;
3467 }
3468 
3469 static bool hns3_get_new_flow_lvl(struct hns3_enet_ring_group *ring_group)
3470 {
3471 #define HNS3_RX_LOW_BYTE_RATE 10000
3472 #define HNS3_RX_MID_BYTE_RATE 20000
3473 #define HNS3_RX_ULTRA_PACKET_RATE 40
3474 
3475 	enum hns3_flow_level_range new_flow_level;
3476 	struct hns3_enet_tqp_vector *tqp_vector;
3477 	int packets_per_msecs, bytes_per_msecs;
3478 	u32 time_passed_ms;
3479 
3480 	tqp_vector = ring_group->ring->tqp_vector;
3481 	time_passed_ms =
3482 		jiffies_to_msecs(jiffies - tqp_vector->last_jiffies);
3483 	if (!time_passed_ms)
3484 		return false;
3485 
3486 	do_div(ring_group->total_packets, time_passed_ms);
3487 	packets_per_msecs = ring_group->total_packets;
3488 
3489 	do_div(ring_group->total_bytes, time_passed_ms);
3490 	bytes_per_msecs = ring_group->total_bytes;
3491 
3492 	new_flow_level = ring_group->coal.flow_level;
3493 
3494 	/* Simple throttlerate management
3495 	 * 0-10MB/s   lower     (50000 ints/s)
3496 	 * 10-20MB/s   middle    (20000 ints/s)
3497 	 * 20-1249MB/s high      (18000 ints/s)
3498 	 * > 40000pps  ultra     (8000 ints/s)
3499 	 */
3500 	switch (new_flow_level) {
3501 	case HNS3_FLOW_LOW:
3502 		if (bytes_per_msecs > HNS3_RX_LOW_BYTE_RATE)
3503 			new_flow_level = HNS3_FLOW_MID;
3504 		break;
3505 	case HNS3_FLOW_MID:
3506 		if (bytes_per_msecs > HNS3_RX_MID_BYTE_RATE)
3507 			new_flow_level = HNS3_FLOW_HIGH;
3508 		else if (bytes_per_msecs <= HNS3_RX_LOW_BYTE_RATE)
3509 			new_flow_level = HNS3_FLOW_LOW;
3510 		break;
3511 	case HNS3_FLOW_HIGH:
3512 	case HNS3_FLOW_ULTRA:
3513 	default:
3514 		if (bytes_per_msecs <= HNS3_RX_MID_BYTE_RATE)
3515 			new_flow_level = HNS3_FLOW_MID;
3516 		break;
3517 	}
3518 
3519 	if (packets_per_msecs > HNS3_RX_ULTRA_PACKET_RATE &&
3520 	    &tqp_vector->rx_group == ring_group)
3521 		new_flow_level = HNS3_FLOW_ULTRA;
3522 
3523 	ring_group->total_bytes = 0;
3524 	ring_group->total_packets = 0;
3525 	ring_group->coal.flow_level = new_flow_level;
3526 
3527 	return true;
3528 }
3529 
3530 static bool hns3_get_new_int_gl(struct hns3_enet_ring_group *ring_group)
3531 {
3532 	struct hns3_enet_tqp_vector *tqp_vector;
3533 	u16 new_int_gl;
3534 
3535 	if (!ring_group->ring)
3536 		return false;
3537 
3538 	tqp_vector = ring_group->ring->tqp_vector;
3539 	if (!tqp_vector->last_jiffies)
3540 		return false;
3541 
3542 	if (ring_group->total_packets == 0) {
3543 		ring_group->coal.int_gl = HNS3_INT_GL_50K;
3544 		ring_group->coal.flow_level = HNS3_FLOW_LOW;
3545 		return true;
3546 	}
3547 
3548 	if (!hns3_get_new_flow_lvl(ring_group))
3549 		return false;
3550 
3551 	new_int_gl = ring_group->coal.int_gl;
3552 	switch (ring_group->coal.flow_level) {
3553 	case HNS3_FLOW_LOW:
3554 		new_int_gl = HNS3_INT_GL_50K;
3555 		break;
3556 	case HNS3_FLOW_MID:
3557 		new_int_gl = HNS3_INT_GL_20K;
3558 		break;
3559 	case HNS3_FLOW_HIGH:
3560 		new_int_gl = HNS3_INT_GL_18K;
3561 		break;
3562 	case HNS3_FLOW_ULTRA:
3563 		new_int_gl = HNS3_INT_GL_8K;
3564 		break;
3565 	default:
3566 		break;
3567 	}
3568 
3569 	if (new_int_gl != ring_group->coal.int_gl) {
3570 		ring_group->coal.int_gl = new_int_gl;
3571 		return true;
3572 	}
3573 	return false;
3574 }
3575 
3576 static void hns3_update_new_int_gl(struct hns3_enet_tqp_vector *tqp_vector)
3577 {
3578 	struct hns3_enet_ring_group *rx_group = &tqp_vector->rx_group;
3579 	struct hns3_enet_ring_group *tx_group = &tqp_vector->tx_group;
3580 	bool rx_update, tx_update;
3581 
3582 	/* update param every 1000ms */
3583 	if (time_before(jiffies,
3584 			tqp_vector->last_jiffies + msecs_to_jiffies(1000)))
3585 		return;
3586 
3587 	if (rx_group->coal.adapt_enable) {
3588 		rx_update = hns3_get_new_int_gl(rx_group);
3589 		if (rx_update)
3590 			hns3_set_vector_coalesce_rx_gl(tqp_vector,
3591 						       rx_group->coal.int_gl);
3592 	}
3593 
3594 	if (tx_group->coal.adapt_enable) {
3595 		tx_update = hns3_get_new_int_gl(tx_group);
3596 		if (tx_update)
3597 			hns3_set_vector_coalesce_tx_gl(tqp_vector,
3598 						       tx_group->coal.int_gl);
3599 	}
3600 
3601 	tqp_vector->last_jiffies = jiffies;
3602 }
3603 
3604 static int hns3_nic_common_poll(struct napi_struct *napi, int budget)
3605 {
3606 	struct hns3_nic_priv *priv = netdev_priv(napi->dev);
3607 	struct hns3_enet_ring *ring;
3608 	int rx_pkt_total = 0;
3609 
3610 	struct hns3_enet_tqp_vector *tqp_vector =
3611 		container_of(napi, struct hns3_enet_tqp_vector, napi);
3612 	bool clean_complete = true;
3613 	int rx_budget = budget;
3614 
3615 	if (unlikely(test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3616 		napi_complete(napi);
3617 		return 0;
3618 	}
3619 
3620 	/* Since the actual Tx work is minimal, we can give the Tx a larger
3621 	 * budget and be more aggressive about cleaning up the Tx descriptors.
3622 	 */
3623 	hns3_for_each_ring(ring, tqp_vector->tx_group)
3624 		hns3_clean_tx_ring(ring, budget);
3625 
3626 	/* make sure rx ring budget not smaller than 1 */
3627 	if (tqp_vector->num_tqps > 1)
3628 		rx_budget = max(budget / tqp_vector->num_tqps, 1);
3629 
3630 	hns3_for_each_ring(ring, tqp_vector->rx_group) {
3631 		int rx_cleaned = hns3_clean_rx_ring(ring, rx_budget,
3632 						    hns3_rx_skb);
3633 		if (rx_cleaned >= rx_budget)
3634 			clean_complete = false;
3635 
3636 		rx_pkt_total += rx_cleaned;
3637 	}
3638 
3639 	tqp_vector->rx_group.total_packets += rx_pkt_total;
3640 
3641 	if (!clean_complete)
3642 		return budget;
3643 
3644 	if (napi_complete(napi) &&
3645 	    likely(!test_bit(HNS3_NIC_STATE_DOWN, &priv->state))) {
3646 		hns3_update_new_int_gl(tqp_vector);
3647 		hns3_mask_vector_irq(tqp_vector, 1);
3648 	}
3649 
3650 	return rx_pkt_total;
3651 }
3652 
3653 static int hns3_get_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3654 				      struct hnae3_ring_chain_node *head)
3655 {
3656 	struct pci_dev *pdev = tqp_vector->handle->pdev;
3657 	struct hnae3_ring_chain_node *cur_chain = head;
3658 	struct hnae3_ring_chain_node *chain;
3659 	struct hns3_enet_ring *tx_ring;
3660 	struct hns3_enet_ring *rx_ring;
3661 
3662 	tx_ring = tqp_vector->tx_group.ring;
3663 	if (tx_ring) {
3664 		cur_chain->tqp_index = tx_ring->tqp->tqp_index;
3665 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3666 			      HNAE3_RING_TYPE_TX);
3667 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3668 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_TX);
3669 
3670 		cur_chain->next = NULL;
3671 
3672 		while (tx_ring->next) {
3673 			tx_ring = tx_ring->next;
3674 
3675 			chain = devm_kzalloc(&pdev->dev, sizeof(*chain),
3676 					     GFP_KERNEL);
3677 			if (!chain)
3678 				goto err_free_chain;
3679 
3680 			cur_chain->next = chain;
3681 			chain->tqp_index = tx_ring->tqp->tqp_index;
3682 			hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3683 				      HNAE3_RING_TYPE_TX);
3684 			hnae3_set_field(chain->int_gl_idx,
3685 					HNAE3_RING_GL_IDX_M,
3686 					HNAE3_RING_GL_IDX_S,
3687 					HNAE3_RING_GL_TX);
3688 
3689 			cur_chain = chain;
3690 		}
3691 	}
3692 
3693 	rx_ring = tqp_vector->rx_group.ring;
3694 	if (!tx_ring && rx_ring) {
3695 		cur_chain->next = NULL;
3696 		cur_chain->tqp_index = rx_ring->tqp->tqp_index;
3697 		hnae3_set_bit(cur_chain->flag, HNAE3_RING_TYPE_B,
3698 			      HNAE3_RING_TYPE_RX);
3699 		hnae3_set_field(cur_chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3700 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3701 
3702 		rx_ring = rx_ring->next;
3703 	}
3704 
3705 	while (rx_ring) {
3706 		chain = devm_kzalloc(&pdev->dev, sizeof(*chain), GFP_KERNEL);
3707 		if (!chain)
3708 			goto err_free_chain;
3709 
3710 		cur_chain->next = chain;
3711 		chain->tqp_index = rx_ring->tqp->tqp_index;
3712 		hnae3_set_bit(chain->flag, HNAE3_RING_TYPE_B,
3713 			      HNAE3_RING_TYPE_RX);
3714 		hnae3_set_field(chain->int_gl_idx, HNAE3_RING_GL_IDX_M,
3715 				HNAE3_RING_GL_IDX_S, HNAE3_RING_GL_RX);
3716 
3717 		cur_chain = chain;
3718 
3719 		rx_ring = rx_ring->next;
3720 	}
3721 
3722 	return 0;
3723 
3724 err_free_chain:
3725 	cur_chain = head->next;
3726 	while (cur_chain) {
3727 		chain = cur_chain->next;
3728 		devm_kfree(&pdev->dev, cur_chain);
3729 		cur_chain = chain;
3730 	}
3731 	head->next = NULL;
3732 
3733 	return -ENOMEM;
3734 }
3735 
3736 static void hns3_free_vector_ring_chain(struct hns3_enet_tqp_vector *tqp_vector,
3737 					struct hnae3_ring_chain_node *head)
3738 {
3739 	struct pci_dev *pdev = tqp_vector->handle->pdev;
3740 	struct hnae3_ring_chain_node *chain_tmp, *chain;
3741 
3742 	chain = head->next;
3743 
3744 	while (chain) {
3745 		chain_tmp = chain->next;
3746 		devm_kfree(&pdev->dev, chain);
3747 		chain = chain_tmp;
3748 	}
3749 }
3750 
3751 static void hns3_add_ring_to_group(struct hns3_enet_ring_group *group,
3752 				   struct hns3_enet_ring *ring)
3753 {
3754 	ring->next = group->ring;
3755 	group->ring = ring;
3756 
3757 	group->count++;
3758 }
3759 
3760 static void hns3_nic_set_cpumask(struct hns3_nic_priv *priv)
3761 {
3762 	struct pci_dev *pdev = priv->ae_handle->pdev;
3763 	struct hns3_enet_tqp_vector *tqp_vector;
3764 	int num_vectors = priv->vector_num;
3765 	int numa_node;
3766 	int vector_i;
3767 
3768 	numa_node = dev_to_node(&pdev->dev);
3769 
3770 	for (vector_i = 0; vector_i < num_vectors; vector_i++) {
3771 		tqp_vector = &priv->tqp_vector[vector_i];
3772 		cpumask_set_cpu(cpumask_local_spread(vector_i, numa_node),
3773 				&tqp_vector->affinity_mask);
3774 	}
3775 }
3776 
3777 static int hns3_nic_init_vector_data(struct hns3_nic_priv *priv)
3778 {
3779 	struct hnae3_handle *h = priv->ae_handle;
3780 	struct hns3_enet_tqp_vector *tqp_vector;
3781 	int ret;
3782 	int i;
3783 
3784 	hns3_nic_set_cpumask(priv);
3785 
3786 	for (i = 0; i < priv->vector_num; i++) {
3787 		tqp_vector = &priv->tqp_vector[i];
3788 		hns3_vector_coalesce_init_hw(tqp_vector, priv);
3789 		tqp_vector->num_tqps = 0;
3790 	}
3791 
3792 	for (i = 0; i < h->kinfo.num_tqps; i++) {
3793 		u16 vector_i = i % priv->vector_num;
3794 		u16 tqp_num = h->kinfo.num_tqps;
3795 
3796 		tqp_vector = &priv->tqp_vector[vector_i];
3797 
3798 		hns3_add_ring_to_group(&tqp_vector->tx_group,
3799 				       &priv->ring[i]);
3800 
3801 		hns3_add_ring_to_group(&tqp_vector->rx_group,
3802 				       &priv->ring[i + tqp_num]);
3803 
3804 		priv->ring[i].tqp_vector = tqp_vector;
3805 		priv->ring[i + tqp_num].tqp_vector = tqp_vector;
3806 		tqp_vector->num_tqps++;
3807 	}
3808 
3809 	for (i = 0; i < priv->vector_num; i++) {
3810 		struct hnae3_ring_chain_node vector_ring_chain;
3811 
3812 		tqp_vector = &priv->tqp_vector[i];
3813 
3814 		tqp_vector->rx_group.total_bytes = 0;
3815 		tqp_vector->rx_group.total_packets = 0;
3816 		tqp_vector->tx_group.total_bytes = 0;
3817 		tqp_vector->tx_group.total_packets = 0;
3818 		tqp_vector->handle = h;
3819 
3820 		ret = hns3_get_vector_ring_chain(tqp_vector,
3821 						 &vector_ring_chain);
3822 		if (ret)
3823 			goto map_ring_fail;
3824 
3825 		ret = h->ae_algo->ops->map_ring_to_vector(h,
3826 			tqp_vector->vector_irq, &vector_ring_chain);
3827 
3828 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3829 
3830 		if (ret)
3831 			goto map_ring_fail;
3832 
3833 		netif_napi_add(priv->netdev, &tqp_vector->napi,
3834 			       hns3_nic_common_poll, NAPI_POLL_WEIGHT);
3835 	}
3836 
3837 	return 0;
3838 
3839 map_ring_fail:
3840 	while (i--)
3841 		netif_napi_del(&priv->tqp_vector[i].napi);
3842 
3843 	return ret;
3844 }
3845 
3846 static int hns3_nic_alloc_vector_data(struct hns3_nic_priv *priv)
3847 {
3848 	struct hnae3_handle *h = priv->ae_handle;
3849 	struct hns3_enet_tqp_vector *tqp_vector;
3850 	struct hnae3_vector_info *vector;
3851 	struct pci_dev *pdev = h->pdev;
3852 	u16 tqp_num = h->kinfo.num_tqps;
3853 	u16 vector_num;
3854 	int ret = 0;
3855 	u16 i;
3856 
3857 	/* RSS size, cpu online and vector_num should be the same */
3858 	/* Should consider 2p/4p later */
3859 	vector_num = min_t(u16, num_online_cpus(), tqp_num);
3860 
3861 	vector = devm_kcalloc(&pdev->dev, vector_num, sizeof(*vector),
3862 			      GFP_KERNEL);
3863 	if (!vector)
3864 		return -ENOMEM;
3865 
3866 	/* save the actual available vector number */
3867 	vector_num = h->ae_algo->ops->get_vector(h, vector_num, vector);
3868 
3869 	priv->vector_num = vector_num;
3870 	priv->tqp_vector = (struct hns3_enet_tqp_vector *)
3871 		devm_kcalloc(&pdev->dev, vector_num, sizeof(*priv->tqp_vector),
3872 			     GFP_KERNEL);
3873 	if (!priv->tqp_vector) {
3874 		ret = -ENOMEM;
3875 		goto out;
3876 	}
3877 
3878 	for (i = 0; i < priv->vector_num; i++) {
3879 		tqp_vector = &priv->tqp_vector[i];
3880 		tqp_vector->idx = i;
3881 		tqp_vector->mask_addr = vector[i].io_addr;
3882 		tqp_vector->vector_irq = vector[i].vector;
3883 		hns3_vector_coalesce_init(tqp_vector, priv);
3884 	}
3885 
3886 out:
3887 	devm_kfree(&pdev->dev, vector);
3888 	return ret;
3889 }
3890 
3891 static void hns3_clear_ring_group(struct hns3_enet_ring_group *group)
3892 {
3893 	group->ring = NULL;
3894 	group->count = 0;
3895 }
3896 
3897 static void hns3_nic_uninit_vector_data(struct hns3_nic_priv *priv)
3898 {
3899 	struct hnae3_ring_chain_node vector_ring_chain;
3900 	struct hnae3_handle *h = priv->ae_handle;
3901 	struct hns3_enet_tqp_vector *tqp_vector;
3902 	int i;
3903 
3904 	for (i = 0; i < priv->vector_num; i++) {
3905 		tqp_vector = &priv->tqp_vector[i];
3906 
3907 		if (!tqp_vector->rx_group.ring && !tqp_vector->tx_group.ring)
3908 			continue;
3909 
3910 		/* Since the mapping can be overwritten, when fail to get the
3911 		 * chain between vector and ring, we should go on to deal with
3912 		 * the remaining options.
3913 		 */
3914 		if (hns3_get_vector_ring_chain(tqp_vector, &vector_ring_chain))
3915 			dev_warn(priv->dev, "failed to get ring chain\n");
3916 
3917 		h->ae_algo->ops->unmap_ring_from_vector(h,
3918 			tqp_vector->vector_irq, &vector_ring_chain);
3919 
3920 		hns3_free_vector_ring_chain(tqp_vector, &vector_ring_chain);
3921 
3922 		hns3_clear_ring_group(&tqp_vector->rx_group);
3923 		hns3_clear_ring_group(&tqp_vector->tx_group);
3924 		netif_napi_del(&priv->tqp_vector[i].napi);
3925 	}
3926 }
3927 
3928 static void hns3_nic_dealloc_vector_data(struct hns3_nic_priv *priv)
3929 {
3930 	struct hnae3_handle *h = priv->ae_handle;
3931 	struct pci_dev *pdev = h->pdev;
3932 	int i, ret;
3933 
3934 	for (i = 0; i < priv->vector_num; i++) {
3935 		struct hns3_enet_tqp_vector *tqp_vector;
3936 
3937 		tqp_vector = &priv->tqp_vector[i];
3938 		ret = h->ae_algo->ops->put_vector(h, tqp_vector->vector_irq);
3939 		if (ret)
3940 			return;
3941 	}
3942 
3943 	devm_kfree(&pdev->dev, priv->tqp_vector);
3944 }
3945 
3946 static void hns3_ring_get_cfg(struct hnae3_queue *q, struct hns3_nic_priv *priv,
3947 			      unsigned int ring_type)
3948 {
3949 	int queue_num = priv->ae_handle->kinfo.num_tqps;
3950 	struct hns3_enet_ring *ring;
3951 	int desc_num;
3952 
3953 	if (ring_type == HNAE3_RING_TYPE_TX) {
3954 		ring = &priv->ring[q->tqp_index];
3955 		desc_num = priv->ae_handle->kinfo.num_tx_desc;
3956 		ring->queue_index = q->tqp_index;
3957 	} else {
3958 		ring = &priv->ring[q->tqp_index + queue_num];
3959 		desc_num = priv->ae_handle->kinfo.num_rx_desc;
3960 		ring->queue_index = q->tqp_index;
3961 	}
3962 
3963 	hnae3_set_bit(ring->flag, HNAE3_RING_TYPE_B, ring_type);
3964 
3965 	ring->tqp = q;
3966 	ring->desc = NULL;
3967 	ring->desc_cb = NULL;
3968 	ring->dev = priv->dev;
3969 	ring->desc_dma_addr = 0;
3970 	ring->buf_size = q->buf_size;
3971 	ring->desc_num = desc_num;
3972 	ring->next_to_use = 0;
3973 	ring->next_to_clean = 0;
3974 	ring->last_to_use = 0;
3975 }
3976 
3977 static void hns3_queue_to_ring(struct hnae3_queue *tqp,
3978 			       struct hns3_nic_priv *priv)
3979 {
3980 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_TX);
3981 	hns3_ring_get_cfg(tqp, priv, HNAE3_RING_TYPE_RX);
3982 }
3983 
3984 static int hns3_get_ring_config(struct hns3_nic_priv *priv)
3985 {
3986 	struct hnae3_handle *h = priv->ae_handle;
3987 	struct pci_dev *pdev = h->pdev;
3988 	int i;
3989 
3990 	priv->ring = devm_kzalloc(&pdev->dev,
3991 				  array3_size(h->kinfo.num_tqps,
3992 					      sizeof(*priv->ring), 2),
3993 				  GFP_KERNEL);
3994 	if (!priv->ring)
3995 		return -ENOMEM;
3996 
3997 	for (i = 0; i < h->kinfo.num_tqps; i++)
3998 		hns3_queue_to_ring(h->kinfo.tqp[i], priv);
3999 
4000 	return 0;
4001 }
4002 
4003 static void hns3_put_ring_config(struct hns3_nic_priv *priv)
4004 {
4005 	if (!priv->ring)
4006 		return;
4007 
4008 	devm_kfree(priv->dev, priv->ring);
4009 	priv->ring = NULL;
4010 }
4011 
4012 static int hns3_alloc_ring_memory(struct hns3_enet_ring *ring)
4013 {
4014 	int ret;
4015 
4016 	if (ring->desc_num <= 0 || ring->buf_size <= 0)
4017 		return -EINVAL;
4018 
4019 	ring->desc_cb = devm_kcalloc(ring_to_dev(ring), ring->desc_num,
4020 				     sizeof(ring->desc_cb[0]), GFP_KERNEL);
4021 	if (!ring->desc_cb) {
4022 		ret = -ENOMEM;
4023 		goto out;
4024 	}
4025 
4026 	ret = hns3_alloc_desc(ring);
4027 	if (ret)
4028 		goto out_with_desc_cb;
4029 
4030 	if (!HNAE3_IS_TX_RING(ring)) {
4031 		ret = hns3_alloc_ring_buffers(ring);
4032 		if (ret)
4033 			goto out_with_desc;
4034 	}
4035 
4036 	return 0;
4037 
4038 out_with_desc:
4039 	hns3_free_desc(ring);
4040 out_with_desc_cb:
4041 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4042 	ring->desc_cb = NULL;
4043 out:
4044 	return ret;
4045 }
4046 
4047 void hns3_fini_ring(struct hns3_enet_ring *ring)
4048 {
4049 	hns3_free_desc(ring);
4050 	devm_kfree(ring_to_dev(ring), ring->desc_cb);
4051 	ring->desc_cb = NULL;
4052 	ring->next_to_clean = 0;
4053 	ring->next_to_use = 0;
4054 	ring->last_to_use = 0;
4055 	ring->pending_buf = 0;
4056 	if (ring->skb) {
4057 		dev_kfree_skb_any(ring->skb);
4058 		ring->skb = NULL;
4059 	}
4060 }
4061 
4062 static int hns3_buf_size2type(u32 buf_size)
4063 {
4064 	int bd_size_type;
4065 
4066 	switch (buf_size) {
4067 	case 512:
4068 		bd_size_type = HNS3_BD_SIZE_512_TYPE;
4069 		break;
4070 	case 1024:
4071 		bd_size_type = HNS3_BD_SIZE_1024_TYPE;
4072 		break;
4073 	case 2048:
4074 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4075 		break;
4076 	case 4096:
4077 		bd_size_type = HNS3_BD_SIZE_4096_TYPE;
4078 		break;
4079 	default:
4080 		bd_size_type = HNS3_BD_SIZE_2048_TYPE;
4081 	}
4082 
4083 	return bd_size_type;
4084 }
4085 
4086 static void hns3_init_ring_hw(struct hns3_enet_ring *ring)
4087 {
4088 	dma_addr_t dma = ring->desc_dma_addr;
4089 	struct hnae3_queue *q = ring->tqp;
4090 
4091 	if (!HNAE3_IS_TX_RING(ring)) {
4092 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_L_REG, (u32)dma);
4093 		hns3_write_dev(q, HNS3_RING_RX_RING_BASEADDR_H_REG,
4094 			       (u32)((dma >> 31) >> 1));
4095 
4096 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_LEN_REG,
4097 			       hns3_buf_size2type(ring->buf_size));
4098 		hns3_write_dev(q, HNS3_RING_RX_RING_BD_NUM_REG,
4099 			       ring->desc_num / 8 - 1);
4100 	} else {
4101 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_L_REG,
4102 			       (u32)dma);
4103 		hns3_write_dev(q, HNS3_RING_TX_RING_BASEADDR_H_REG,
4104 			       (u32)((dma >> 31) >> 1));
4105 
4106 		hns3_write_dev(q, HNS3_RING_TX_RING_BD_NUM_REG,
4107 			       ring->desc_num / 8 - 1);
4108 	}
4109 }
4110 
4111 static void hns3_init_tx_ring_tc(struct hns3_nic_priv *priv)
4112 {
4113 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4114 	struct hnae3_tc_info *tc_info = &kinfo->tc_info;
4115 	int i;
4116 
4117 	for (i = 0; i < HNAE3_MAX_TC; i++) {
4118 		int j;
4119 
4120 		if (!test_bit(i, &tc_info->tc_en))
4121 			continue;
4122 
4123 		for (j = 0; j < tc_info->tqp_count[i]; j++) {
4124 			struct hnae3_queue *q;
4125 
4126 			q = priv->ring[tc_info->tqp_offset[i] + j].tqp;
4127 			hns3_write_dev(q, HNS3_RING_TX_RING_TC_REG, i);
4128 		}
4129 	}
4130 }
4131 
4132 int hns3_init_all_ring(struct hns3_nic_priv *priv)
4133 {
4134 	struct hnae3_handle *h = priv->ae_handle;
4135 	int ring_num = h->kinfo.num_tqps * 2;
4136 	int i, j;
4137 	int ret;
4138 
4139 	for (i = 0; i < ring_num; i++) {
4140 		ret = hns3_alloc_ring_memory(&priv->ring[i]);
4141 		if (ret) {
4142 			dev_err(priv->dev,
4143 				"Alloc ring memory fail! ret=%d\n", ret);
4144 			goto out_when_alloc_ring_memory;
4145 		}
4146 
4147 		u64_stats_init(&priv->ring[i].syncp);
4148 	}
4149 
4150 	return 0;
4151 
4152 out_when_alloc_ring_memory:
4153 	for (j = i - 1; j >= 0; j--)
4154 		hns3_fini_ring(&priv->ring[j]);
4155 
4156 	return -ENOMEM;
4157 }
4158 
4159 static void hns3_uninit_all_ring(struct hns3_nic_priv *priv)
4160 {
4161 	struct hnae3_handle *h = priv->ae_handle;
4162 	int i;
4163 
4164 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4165 		hns3_fini_ring(&priv->ring[i]);
4166 		hns3_fini_ring(&priv->ring[i + h->kinfo.num_tqps]);
4167 	}
4168 }
4169 
4170 /* Set mac addr if it is configured. or leave it to the AE driver */
4171 static int hns3_init_mac_addr(struct net_device *netdev)
4172 {
4173 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4174 	struct hnae3_handle *h = priv->ae_handle;
4175 	u8 mac_addr_temp[ETH_ALEN];
4176 	int ret = 0;
4177 
4178 	if (h->ae_algo->ops->get_mac_addr)
4179 		h->ae_algo->ops->get_mac_addr(h, mac_addr_temp);
4180 
4181 	/* Check if the MAC address is valid, if not get a random one */
4182 	if (!is_valid_ether_addr(mac_addr_temp)) {
4183 		eth_hw_addr_random(netdev);
4184 		dev_warn(priv->dev, "using random MAC address %pM\n",
4185 			 netdev->dev_addr);
4186 	} else if (!ether_addr_equal(netdev->dev_addr, mac_addr_temp)) {
4187 		ether_addr_copy(netdev->dev_addr, mac_addr_temp);
4188 		ether_addr_copy(netdev->perm_addr, mac_addr_temp);
4189 	} else {
4190 		return 0;
4191 	}
4192 
4193 	if (h->ae_algo->ops->set_mac_addr)
4194 		ret = h->ae_algo->ops->set_mac_addr(h, netdev->dev_addr, true);
4195 
4196 	return ret;
4197 }
4198 
4199 static int hns3_init_phy(struct net_device *netdev)
4200 {
4201 	struct hnae3_handle *h = hns3_get_handle(netdev);
4202 	int ret = 0;
4203 
4204 	if (h->ae_algo->ops->mac_connect_phy)
4205 		ret = h->ae_algo->ops->mac_connect_phy(h);
4206 
4207 	return ret;
4208 }
4209 
4210 static void hns3_uninit_phy(struct net_device *netdev)
4211 {
4212 	struct hnae3_handle *h = hns3_get_handle(netdev);
4213 
4214 	if (h->ae_algo->ops->mac_disconnect_phy)
4215 		h->ae_algo->ops->mac_disconnect_phy(h);
4216 }
4217 
4218 static int hns3_client_start(struct hnae3_handle *handle)
4219 {
4220 	if (!handle->ae_algo->ops->client_start)
4221 		return 0;
4222 
4223 	return handle->ae_algo->ops->client_start(handle);
4224 }
4225 
4226 static void hns3_client_stop(struct hnae3_handle *handle)
4227 {
4228 	if (!handle->ae_algo->ops->client_stop)
4229 		return;
4230 
4231 	handle->ae_algo->ops->client_stop(handle);
4232 }
4233 
4234 static void hns3_info_show(struct hns3_nic_priv *priv)
4235 {
4236 	struct hnae3_knic_private_info *kinfo = &priv->ae_handle->kinfo;
4237 
4238 	dev_info(priv->dev, "MAC address: %pM\n", priv->netdev->dev_addr);
4239 	dev_info(priv->dev, "Task queue pairs numbers: %u\n", kinfo->num_tqps);
4240 	dev_info(priv->dev, "RSS size: %u\n", kinfo->rss_size);
4241 	dev_info(priv->dev, "Allocated RSS size: %u\n", kinfo->req_rss_size);
4242 	dev_info(priv->dev, "RX buffer length: %u\n", kinfo->rx_buf_len);
4243 	dev_info(priv->dev, "Desc num per TX queue: %u\n", kinfo->num_tx_desc);
4244 	dev_info(priv->dev, "Desc num per RX queue: %u\n", kinfo->num_rx_desc);
4245 	dev_info(priv->dev, "Total number of enabled TCs: %u\n",
4246 		 kinfo->tc_info.num_tc);
4247 	dev_info(priv->dev, "Max mtu size: %u\n", priv->netdev->max_mtu);
4248 }
4249 
4250 static int hns3_client_init(struct hnae3_handle *handle)
4251 {
4252 	struct pci_dev *pdev = handle->pdev;
4253 	struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
4254 	u16 alloc_tqps, max_rss_size;
4255 	struct hns3_nic_priv *priv;
4256 	struct net_device *netdev;
4257 	int ret;
4258 
4259 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
4260 						    &max_rss_size);
4261 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
4262 	if (!netdev)
4263 		return -ENOMEM;
4264 
4265 	priv = netdev_priv(netdev);
4266 	priv->dev = &pdev->dev;
4267 	priv->netdev = netdev;
4268 	priv->ae_handle = handle;
4269 	priv->tx_timeout_count = 0;
4270 	priv->max_non_tso_bd_num = ae_dev->dev_specs.max_non_tso_bd_num;
4271 	set_bit(HNS3_NIC_STATE_DOWN, &priv->state);
4272 
4273 	handle->msg_enable = netif_msg_init(debug, DEFAULT_MSG_LEVEL);
4274 
4275 	handle->kinfo.netdev = netdev;
4276 	handle->priv = (void *)priv;
4277 
4278 	hns3_init_mac_addr(netdev);
4279 
4280 	hns3_set_default_feature(netdev);
4281 
4282 	netdev->watchdog_timeo = HNS3_TX_TIMEOUT;
4283 	netdev->priv_flags |= IFF_UNICAST_FLT;
4284 	netdev->netdev_ops = &hns3_nic_netdev_ops;
4285 	SET_NETDEV_DEV(netdev, &pdev->dev);
4286 	hns3_ethtool_set_ops(netdev);
4287 
4288 	/* Carrier off reporting is important to ethtool even BEFORE open */
4289 	netif_carrier_off(netdev);
4290 
4291 	ret = hns3_get_ring_config(priv);
4292 	if (ret) {
4293 		ret = -ENOMEM;
4294 		goto out_get_ring_cfg;
4295 	}
4296 
4297 	ret = hns3_nic_alloc_vector_data(priv);
4298 	if (ret) {
4299 		ret = -ENOMEM;
4300 		goto out_alloc_vector_data;
4301 	}
4302 
4303 	ret = hns3_nic_init_vector_data(priv);
4304 	if (ret) {
4305 		ret = -ENOMEM;
4306 		goto out_init_vector_data;
4307 	}
4308 
4309 	ret = hns3_init_all_ring(priv);
4310 	if (ret) {
4311 		ret = -ENOMEM;
4312 		goto out_init_ring;
4313 	}
4314 
4315 	ret = hns3_init_phy(netdev);
4316 	if (ret)
4317 		goto out_init_phy;
4318 
4319 	ret = register_netdev(netdev);
4320 	if (ret) {
4321 		dev_err(priv->dev, "probe register netdev fail!\n");
4322 		goto out_reg_netdev_fail;
4323 	}
4324 
4325 	/* the device can work without cpu rmap, only aRFS needs it */
4326 	ret = hns3_set_rx_cpu_rmap(netdev);
4327 	if (ret)
4328 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4329 
4330 	ret = hns3_nic_init_irq(priv);
4331 	if (ret) {
4332 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4333 		hns3_free_rx_cpu_rmap(netdev);
4334 		goto out_init_irq_fail;
4335 	}
4336 
4337 	ret = hns3_client_start(handle);
4338 	if (ret) {
4339 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4340 		goto out_client_start;
4341 	}
4342 
4343 	hns3_dcbnl_setup(handle);
4344 
4345 	hns3_dbg_init(handle);
4346 
4347 	netdev->max_mtu = HNS3_MAX_MTU(ae_dev->dev_specs.max_frm_size);
4348 
4349 	if (test_bit(HNAE3_DEV_SUPPORT_HW_TX_CSUM_B, ae_dev->caps))
4350 		set_bit(HNS3_NIC_STATE_HW_TX_CSUM_ENABLE, &priv->state);
4351 
4352 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4353 
4354 	if (ae_dev->dev_version >= HNAE3_DEVICE_VERSION_V3)
4355 		set_bit(HNAE3_PFLAG_LIMIT_PROMISC, &handle->supported_pflags);
4356 
4357 	if (netif_msg_drv(handle))
4358 		hns3_info_show(priv);
4359 
4360 	return ret;
4361 
4362 out_client_start:
4363 	hns3_free_rx_cpu_rmap(netdev);
4364 	hns3_nic_uninit_irq(priv);
4365 out_init_irq_fail:
4366 	unregister_netdev(netdev);
4367 out_reg_netdev_fail:
4368 	hns3_uninit_phy(netdev);
4369 out_init_phy:
4370 	hns3_uninit_all_ring(priv);
4371 out_init_ring:
4372 	hns3_nic_uninit_vector_data(priv);
4373 out_init_vector_data:
4374 	hns3_nic_dealloc_vector_data(priv);
4375 out_alloc_vector_data:
4376 	priv->ring = NULL;
4377 out_get_ring_cfg:
4378 	priv->ae_handle = NULL;
4379 	free_netdev(netdev);
4380 	return ret;
4381 }
4382 
4383 static void hns3_client_uninit(struct hnae3_handle *handle, bool reset)
4384 {
4385 	struct net_device *netdev = handle->kinfo.netdev;
4386 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4387 
4388 	if (netdev->reg_state != NETREG_UNINITIALIZED)
4389 		unregister_netdev(netdev);
4390 
4391 	hns3_client_stop(handle);
4392 
4393 	hns3_uninit_phy(netdev);
4394 
4395 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4396 		netdev_warn(netdev, "already uninitialized\n");
4397 		goto out_netdev_free;
4398 	}
4399 
4400 	hns3_free_rx_cpu_rmap(netdev);
4401 
4402 	hns3_nic_uninit_irq(priv);
4403 
4404 	hns3_clear_all_ring(handle, true);
4405 
4406 	hns3_nic_uninit_vector_data(priv);
4407 
4408 	hns3_nic_dealloc_vector_data(priv);
4409 
4410 	hns3_uninit_all_ring(priv);
4411 
4412 	hns3_put_ring_config(priv);
4413 
4414 out_netdev_free:
4415 	hns3_dbg_uninit(handle);
4416 	free_netdev(netdev);
4417 }
4418 
4419 static void hns3_link_status_change(struct hnae3_handle *handle, bool linkup)
4420 {
4421 	struct net_device *netdev = handle->kinfo.netdev;
4422 
4423 	if (!netdev)
4424 		return;
4425 
4426 	if (linkup) {
4427 		netif_tx_wake_all_queues(netdev);
4428 		netif_carrier_on(netdev);
4429 		if (netif_msg_link(handle))
4430 			netdev_info(netdev, "link up\n");
4431 	} else {
4432 		netif_carrier_off(netdev);
4433 		netif_tx_stop_all_queues(netdev);
4434 		if (netif_msg_link(handle))
4435 			netdev_info(netdev, "link down\n");
4436 	}
4437 }
4438 
4439 static void hns3_clear_tx_ring(struct hns3_enet_ring *ring)
4440 {
4441 	while (ring->next_to_clean != ring->next_to_use) {
4442 		ring->desc[ring->next_to_clean].tx.bdtp_fe_sc_vld_ra_ri = 0;
4443 		hns3_free_buffer_detach(ring, ring->next_to_clean, 0);
4444 		ring_ptr_move_fw(ring, next_to_clean);
4445 	}
4446 
4447 	ring->pending_buf = 0;
4448 }
4449 
4450 static int hns3_clear_rx_ring(struct hns3_enet_ring *ring)
4451 {
4452 	struct hns3_desc_cb res_cbs;
4453 	int ret;
4454 
4455 	while (ring->next_to_use != ring->next_to_clean) {
4456 		/* When a buffer is not reused, it's memory has been
4457 		 * freed in hns3_handle_rx_bd or will be freed by
4458 		 * stack, so we need to replace the buffer here.
4459 		 */
4460 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4461 			ret = hns3_alloc_and_map_buffer(ring, &res_cbs);
4462 			if (ret) {
4463 				u64_stats_update_begin(&ring->syncp);
4464 				ring->stats.sw_err_cnt++;
4465 				u64_stats_update_end(&ring->syncp);
4466 				/* if alloc new buffer fail, exit directly
4467 				 * and reclear in up flow.
4468 				 */
4469 				netdev_warn(ring_to_netdev(ring),
4470 					    "reserve buffer map failed, ret = %d\n",
4471 					    ret);
4472 				return ret;
4473 			}
4474 			hns3_replace_buffer(ring, ring->next_to_use, &res_cbs);
4475 		}
4476 		ring_ptr_move_fw(ring, next_to_use);
4477 	}
4478 
4479 	/* Free the pending skb in rx ring */
4480 	if (ring->skb) {
4481 		dev_kfree_skb_any(ring->skb);
4482 		ring->skb = NULL;
4483 		ring->pending_buf = 0;
4484 	}
4485 
4486 	return 0;
4487 }
4488 
4489 static void hns3_force_clear_rx_ring(struct hns3_enet_ring *ring)
4490 {
4491 	while (ring->next_to_use != ring->next_to_clean) {
4492 		/* When a buffer is not reused, it's memory has been
4493 		 * freed in hns3_handle_rx_bd or will be freed by
4494 		 * stack, so only need to unmap the buffer here.
4495 		 */
4496 		if (!ring->desc_cb[ring->next_to_use].reuse_flag) {
4497 			hns3_unmap_buffer(ring,
4498 					  &ring->desc_cb[ring->next_to_use]);
4499 			ring->desc_cb[ring->next_to_use].dma = 0;
4500 		}
4501 
4502 		ring_ptr_move_fw(ring, next_to_use);
4503 	}
4504 }
4505 
4506 static void hns3_clear_all_ring(struct hnae3_handle *h, bool force)
4507 {
4508 	struct net_device *ndev = h->kinfo.netdev;
4509 	struct hns3_nic_priv *priv = netdev_priv(ndev);
4510 	u32 i;
4511 
4512 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4513 		struct hns3_enet_ring *ring;
4514 
4515 		ring = &priv->ring[i];
4516 		hns3_clear_tx_ring(ring);
4517 
4518 		ring = &priv->ring[i + h->kinfo.num_tqps];
4519 		/* Continue to clear other rings even if clearing some
4520 		 * rings failed.
4521 		 */
4522 		if (force)
4523 			hns3_force_clear_rx_ring(ring);
4524 		else
4525 			hns3_clear_rx_ring(ring);
4526 	}
4527 }
4528 
4529 int hns3_nic_reset_all_ring(struct hnae3_handle *h)
4530 {
4531 	struct net_device *ndev = h->kinfo.netdev;
4532 	struct hns3_nic_priv *priv = netdev_priv(ndev);
4533 	struct hns3_enet_ring *rx_ring;
4534 	int i, j;
4535 	int ret;
4536 
4537 	ret = h->ae_algo->ops->reset_queue(h);
4538 	if (ret)
4539 		return ret;
4540 
4541 	for (i = 0; i < h->kinfo.num_tqps; i++) {
4542 		hns3_init_ring_hw(&priv->ring[i]);
4543 
4544 		/* We need to clear tx ring here because self test will
4545 		 * use the ring and will not run down before up
4546 		 */
4547 		hns3_clear_tx_ring(&priv->ring[i]);
4548 		priv->ring[i].next_to_clean = 0;
4549 		priv->ring[i].next_to_use = 0;
4550 		priv->ring[i].last_to_use = 0;
4551 
4552 		rx_ring = &priv->ring[i + h->kinfo.num_tqps];
4553 		hns3_init_ring_hw(rx_ring);
4554 		ret = hns3_clear_rx_ring(rx_ring);
4555 		if (ret)
4556 			return ret;
4557 
4558 		/* We can not know the hardware head and tail when this
4559 		 * function is called in reset flow, so we reuse all desc.
4560 		 */
4561 		for (j = 0; j < rx_ring->desc_num; j++)
4562 			hns3_reuse_buffer(rx_ring, j);
4563 
4564 		rx_ring->next_to_clean = 0;
4565 		rx_ring->next_to_use = 0;
4566 	}
4567 
4568 	hns3_init_tx_ring_tc(priv);
4569 
4570 	return 0;
4571 }
4572 
4573 static void hns3_store_coal(struct hns3_nic_priv *priv)
4574 {
4575 	/* ethtool only support setting and querying one coal
4576 	 * configuration for now, so save the vector 0' coal
4577 	 * configuration here in order to restore it.
4578 	 */
4579 	memcpy(&priv->tx_coal, &priv->tqp_vector[0].tx_group.coal,
4580 	       sizeof(struct hns3_enet_coalesce));
4581 	memcpy(&priv->rx_coal, &priv->tqp_vector[0].rx_group.coal,
4582 	       sizeof(struct hns3_enet_coalesce));
4583 }
4584 
4585 static void hns3_restore_coal(struct hns3_nic_priv *priv)
4586 {
4587 	u16 vector_num = priv->vector_num;
4588 	int i;
4589 
4590 	for (i = 0; i < vector_num; i++) {
4591 		memcpy(&priv->tqp_vector[i].tx_group.coal, &priv->tx_coal,
4592 		       sizeof(struct hns3_enet_coalesce));
4593 		memcpy(&priv->tqp_vector[i].rx_group.coal, &priv->rx_coal,
4594 		       sizeof(struct hns3_enet_coalesce));
4595 	}
4596 }
4597 
4598 static int hns3_reset_notify_down_enet(struct hnae3_handle *handle)
4599 {
4600 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4601 	struct net_device *ndev = kinfo->netdev;
4602 	struct hns3_nic_priv *priv = netdev_priv(ndev);
4603 
4604 	if (test_and_set_bit(HNS3_NIC_STATE_RESETTING, &priv->state))
4605 		return 0;
4606 
4607 	if (!netif_running(ndev))
4608 		return 0;
4609 
4610 	return hns3_nic_net_stop(ndev);
4611 }
4612 
4613 static int hns3_reset_notify_up_enet(struct hnae3_handle *handle)
4614 {
4615 	struct hnae3_knic_private_info *kinfo = &handle->kinfo;
4616 	struct hns3_nic_priv *priv = netdev_priv(kinfo->netdev);
4617 	int ret = 0;
4618 
4619 	clear_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4620 
4621 	if (netif_running(kinfo->netdev)) {
4622 		ret = hns3_nic_net_open(kinfo->netdev);
4623 		if (ret) {
4624 			set_bit(HNS3_NIC_STATE_RESETTING, &priv->state);
4625 			netdev_err(kinfo->netdev,
4626 				   "net up fail, ret=%d!\n", ret);
4627 			return ret;
4628 		}
4629 	}
4630 
4631 	return ret;
4632 }
4633 
4634 static int hns3_reset_notify_init_enet(struct hnae3_handle *handle)
4635 {
4636 	struct net_device *netdev = handle->kinfo.netdev;
4637 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4638 	int ret;
4639 
4640 	/* Carrier off reporting is important to ethtool even BEFORE open */
4641 	netif_carrier_off(netdev);
4642 
4643 	ret = hns3_get_ring_config(priv);
4644 	if (ret)
4645 		return ret;
4646 
4647 	ret = hns3_nic_alloc_vector_data(priv);
4648 	if (ret)
4649 		goto err_put_ring;
4650 
4651 	hns3_restore_coal(priv);
4652 
4653 	ret = hns3_nic_init_vector_data(priv);
4654 	if (ret)
4655 		goto err_dealloc_vector;
4656 
4657 	ret = hns3_init_all_ring(priv);
4658 	if (ret)
4659 		goto err_uninit_vector;
4660 
4661 	/* the device can work without cpu rmap, only aRFS needs it */
4662 	ret = hns3_set_rx_cpu_rmap(netdev);
4663 	if (ret)
4664 		dev_warn(priv->dev, "set rx cpu rmap fail, ret=%d\n", ret);
4665 
4666 	ret = hns3_nic_init_irq(priv);
4667 	if (ret) {
4668 		dev_err(priv->dev, "init irq failed! ret=%d\n", ret);
4669 		hns3_free_rx_cpu_rmap(netdev);
4670 		goto err_init_irq_fail;
4671 	}
4672 
4673 	if (!hns3_is_phys_func(handle->pdev))
4674 		hns3_init_mac_addr(netdev);
4675 
4676 	ret = hns3_client_start(handle);
4677 	if (ret) {
4678 		dev_err(priv->dev, "hns3_client_start fail! ret=%d\n", ret);
4679 		goto err_client_start_fail;
4680 	}
4681 
4682 	set_bit(HNS3_NIC_STATE_INITED, &priv->state);
4683 
4684 	return ret;
4685 
4686 err_client_start_fail:
4687 	hns3_free_rx_cpu_rmap(netdev);
4688 	hns3_nic_uninit_irq(priv);
4689 err_init_irq_fail:
4690 	hns3_uninit_all_ring(priv);
4691 err_uninit_vector:
4692 	hns3_nic_uninit_vector_data(priv);
4693 err_dealloc_vector:
4694 	hns3_nic_dealloc_vector_data(priv);
4695 err_put_ring:
4696 	hns3_put_ring_config(priv);
4697 
4698 	return ret;
4699 }
4700 
4701 static int hns3_reset_notify_uninit_enet(struct hnae3_handle *handle)
4702 {
4703 	struct net_device *netdev = handle->kinfo.netdev;
4704 	struct hns3_nic_priv *priv = netdev_priv(netdev);
4705 
4706 	if (!test_and_clear_bit(HNS3_NIC_STATE_INITED, &priv->state)) {
4707 		netdev_warn(netdev, "already uninitialized\n");
4708 		return 0;
4709 	}
4710 
4711 	hns3_free_rx_cpu_rmap(netdev);
4712 	hns3_nic_uninit_irq(priv);
4713 	hns3_clear_all_ring(handle, true);
4714 	hns3_reset_tx_queue(priv->ae_handle);
4715 
4716 	hns3_nic_uninit_vector_data(priv);
4717 
4718 	hns3_store_coal(priv);
4719 
4720 	hns3_nic_dealloc_vector_data(priv);
4721 
4722 	hns3_uninit_all_ring(priv);
4723 
4724 	hns3_put_ring_config(priv);
4725 
4726 	return 0;
4727 }
4728 
4729 static int hns3_reset_notify(struct hnae3_handle *handle,
4730 			     enum hnae3_reset_notify_type type)
4731 {
4732 	int ret = 0;
4733 
4734 	switch (type) {
4735 	case HNAE3_UP_CLIENT:
4736 		ret = hns3_reset_notify_up_enet(handle);
4737 		break;
4738 	case HNAE3_DOWN_CLIENT:
4739 		ret = hns3_reset_notify_down_enet(handle);
4740 		break;
4741 	case HNAE3_INIT_CLIENT:
4742 		ret = hns3_reset_notify_init_enet(handle);
4743 		break;
4744 	case HNAE3_UNINIT_CLIENT:
4745 		ret = hns3_reset_notify_uninit_enet(handle);
4746 		break;
4747 	default:
4748 		break;
4749 	}
4750 
4751 	return ret;
4752 }
4753 
4754 static int hns3_change_channels(struct hnae3_handle *handle, u32 new_tqp_num,
4755 				bool rxfh_configured)
4756 {
4757 	int ret;
4758 
4759 	ret = handle->ae_algo->ops->set_channels(handle, new_tqp_num,
4760 						 rxfh_configured);
4761 	if (ret) {
4762 		dev_err(&handle->pdev->dev,
4763 			"Change tqp num(%u) fail.\n", new_tqp_num);
4764 		return ret;
4765 	}
4766 
4767 	ret = hns3_reset_notify(handle, HNAE3_INIT_CLIENT);
4768 	if (ret)
4769 		return ret;
4770 
4771 	ret =  hns3_reset_notify(handle, HNAE3_UP_CLIENT);
4772 	if (ret)
4773 		hns3_reset_notify(handle, HNAE3_UNINIT_CLIENT);
4774 
4775 	return ret;
4776 }
4777 
4778 int hns3_set_channels(struct net_device *netdev,
4779 		      struct ethtool_channels *ch)
4780 {
4781 	struct hnae3_handle *h = hns3_get_handle(netdev);
4782 	struct hnae3_knic_private_info *kinfo = &h->kinfo;
4783 	bool rxfh_configured = netif_is_rxfh_configured(netdev);
4784 	u32 new_tqp_num = ch->combined_count;
4785 	u16 org_tqp_num;
4786 	int ret;
4787 
4788 	if (hns3_nic_resetting(netdev))
4789 		return -EBUSY;
4790 
4791 	if (ch->rx_count || ch->tx_count)
4792 		return -EINVAL;
4793 
4794 	if (kinfo->tc_info.mqprio_active) {
4795 		dev_err(&netdev->dev,
4796 			"it's not allowed to set channels via ethtool when MQPRIO mode is on\n");
4797 		return -EINVAL;
4798 	}
4799 
4800 	if (new_tqp_num > hns3_get_max_available_channels(h) ||
4801 	    new_tqp_num < 1) {
4802 		dev_err(&netdev->dev,
4803 			"Change tqps fail, the tqp range is from 1 to %u",
4804 			hns3_get_max_available_channels(h));
4805 		return -EINVAL;
4806 	}
4807 
4808 	if (kinfo->rss_size == new_tqp_num)
4809 		return 0;
4810 
4811 	netif_dbg(h, drv, netdev,
4812 		  "set channels: tqp_num=%u, rxfh=%d\n",
4813 		  new_tqp_num, rxfh_configured);
4814 
4815 	ret = hns3_reset_notify(h, HNAE3_DOWN_CLIENT);
4816 	if (ret)
4817 		return ret;
4818 
4819 	ret = hns3_reset_notify(h, HNAE3_UNINIT_CLIENT);
4820 	if (ret)
4821 		return ret;
4822 
4823 	org_tqp_num = h->kinfo.num_tqps;
4824 	ret = hns3_change_channels(h, new_tqp_num, rxfh_configured);
4825 	if (ret) {
4826 		int ret1;
4827 
4828 		netdev_warn(netdev,
4829 			    "Change channels fail, revert to old value\n");
4830 		ret1 = hns3_change_channels(h, org_tqp_num, rxfh_configured);
4831 		if (ret1) {
4832 			netdev_err(netdev,
4833 				   "revert to old channel fail\n");
4834 			return ret1;
4835 		}
4836 
4837 		return ret;
4838 	}
4839 
4840 	return 0;
4841 }
4842 
4843 static const struct hns3_hw_error_info hns3_hw_err[] = {
4844 	{ .type = HNAE3_PPU_POISON_ERROR,
4845 	  .msg = "PPU poison" },
4846 	{ .type = HNAE3_CMDQ_ECC_ERROR,
4847 	  .msg = "IMP CMDQ error" },
4848 	{ .type = HNAE3_IMP_RD_POISON_ERROR,
4849 	  .msg = "IMP RD poison" },
4850 	{ .type = HNAE3_ROCEE_AXI_RESP_ERROR,
4851 	  .msg = "ROCEE AXI RESP error" },
4852 };
4853 
4854 static void hns3_process_hw_error(struct hnae3_handle *handle,
4855 				  enum hnae3_hw_error_type type)
4856 {
4857 	int i;
4858 
4859 	for (i = 0; i < ARRAY_SIZE(hns3_hw_err); i++) {
4860 		if (hns3_hw_err[i].type == type) {
4861 			dev_err(&handle->pdev->dev, "Detected %s!\n",
4862 				hns3_hw_err[i].msg);
4863 			break;
4864 		}
4865 	}
4866 }
4867 
4868 static const struct hnae3_client_ops client_ops = {
4869 	.init_instance = hns3_client_init,
4870 	.uninit_instance = hns3_client_uninit,
4871 	.link_status_change = hns3_link_status_change,
4872 	.reset_notify = hns3_reset_notify,
4873 	.process_hw_error = hns3_process_hw_error,
4874 };
4875 
4876 /* hns3_init_module - Driver registration routine
4877  * hns3_init_module is the first routine called when the driver is
4878  * loaded. All it does is register with the PCI subsystem.
4879  */
4880 static int __init hns3_init_module(void)
4881 {
4882 	int ret;
4883 
4884 	pr_info("%s: %s - version\n", hns3_driver_name, hns3_driver_string);
4885 	pr_info("%s: %s\n", hns3_driver_name, hns3_copyright);
4886 
4887 	client.type = HNAE3_CLIENT_KNIC;
4888 	snprintf(client.name, HNAE3_CLIENT_NAME_LENGTH, "%s",
4889 		 hns3_driver_name);
4890 
4891 	client.ops = &client_ops;
4892 
4893 	INIT_LIST_HEAD(&client.node);
4894 
4895 	hns3_dbg_register_debugfs(hns3_driver_name);
4896 
4897 	ret = hnae3_register_client(&client);
4898 	if (ret)
4899 		goto err_reg_client;
4900 
4901 	ret = pci_register_driver(&hns3_driver);
4902 	if (ret)
4903 		goto err_reg_driver;
4904 
4905 	return ret;
4906 
4907 err_reg_driver:
4908 	hnae3_unregister_client(&client);
4909 err_reg_client:
4910 	hns3_dbg_unregister_debugfs();
4911 	return ret;
4912 }
4913 module_init(hns3_init_module);
4914 
4915 /* hns3_exit_module - Driver exit cleanup routine
4916  * hns3_exit_module is called just before the driver is removed
4917  * from memory.
4918  */
4919 static void __exit hns3_exit_module(void)
4920 {
4921 	pci_unregister_driver(&hns3_driver);
4922 	hnae3_unregister_client(&client);
4923 	hns3_dbg_unregister_debugfs();
4924 }
4925 module_exit(hns3_exit_module);
4926 
4927 MODULE_DESCRIPTION("HNS3: Hisilicon Ethernet Driver");
4928 MODULE_AUTHOR("Huawei Tech. Co., Ltd.");
4929 MODULE_LICENSE("GPL");
4930 MODULE_ALIAS("pci:hns-nic");
4931